{"id": "train_035000", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a4a07695316be59b2d47e84f81bd8341c70e9257"}
{"id": "train_035001", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a45687fd9c94d4c08703d2a3193974f869db65a2"}
{"id": "train_035002", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "501f66c4fa45b17afb3c438deafafc9d0e57dace"}
{"id": "train_035003", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["evaluation", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "982ea1e2ac47ea46a30011ff6d321ea0fec69934"}
{"id": "train_035004", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "5585eb384ace4a9499c4cc51b3bdbfd93e2d5743"}
{"id": "train_035005", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "a7b6692bb60d639e4c501b95fd7dae36e8cc218b"}
{"id": "train_035006", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c49b5f938762bade659b12969d4f058cccd8fe59"}
{"id": "train_035007", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "b26461888e1c2bde19818eea7f70b9168d3df771"}
{"id": "train_035008", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ec92a39cbfa5c222641807c5b7b231b95ab1ae1b"}
{"id": "train_035009", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "13eb5e9c07b6b20896a37ad65067d0b6ade40579"}
{"id": "train_035010", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "c6feffb0192f86663d18c225a9947452fdd4d5e9"}
{"id": "train_035011", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "79274bed166f7221c6c3f986ed0ec02539fc5fb9"}
{"id": "train_035012", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "a29a68d2bb5c4f20f60b93305f9d262f3c6a03d4"}
{"id": "train_035013", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "8638dfce1e9dde4c4701b122a09c32714be5dc5b"}
{"id": "train_035014", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f35d18504bc0c7e1d4a75bdf0dc72a4bf9b1ce09"}
{"id": "train_035015", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "5d5595d0e6b1edf0b6cacab9d534d876a72b8513"}
{"id": "train_035016", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "a7ec826fe63894d38054484c7befc88ab0290110"}
{"id": "train_035017", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "4de64016b693b04816eaf1aed2c86e64aab56b74"}
{"id": "train_035018", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "70766f104729592af956911c5f6253052d46c1b3"}
{"id": "train_035019", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "eff785af56a4385e33fb37529c1a861a17f846fa"}
{"id": "train_035020", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "ea49fcf5ea93dfd0fe8956b3986cc845fea5f7b5"}
{"id": "train_035021", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["performance", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d19f0dda46923abaa3a26640e35655c64f4f165a"}
{"id": "train_035022", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "07278b24fac8e7f18a9a18d602dace725bb06906"}
{"id": "train_035023", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "68321ed0472f4d85353311473a6c1835d72629f2"}
{"id": "train_035024", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "812b4ffeedad3d3207a75c2ab59fd8e683f9c576"}
{"id": "train_035025", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b3735d3c8f389f10b22a53af27c7fbc852cd1a7e"}
{"id": "train_035026", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "63b2dc2218ee592f1cdd2f99de52862b621144df"}
{"id": "train_035027", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "edc676261e5434ef217ce6f9986ce9550c7611e1"}
{"id": "train_035028", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "321334a26e9dfd74e3f7d8f7bfdd28a6aadaf438"}
{"id": "train_035029", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "1224e837feeffa5520a04594113fd81f1b672f4c"}
{"id": "train_035030", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "8e9fa220a40f255cfeb1c2d9d606fa475a8498df"}
{"id": "train_035031", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c27634e9f36382d3e516056e01b41e5b3b08870f"}
{"id": "train_035032", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "4961cf7dcdeabf2664b7c749c9b7bb69ee1bad6d"}
{"id": "train_035033", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "6434b9c53f6834963e7cab3c9080be5a477eeb6a"}
{"id": "train_035034", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "06a3366d2197a60939fae40a583a456cb480905d"}
{"id": "train_035035", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2bff81d98807c7a456918b33d732c52744c86471"}
{"id": "train_035036", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "817067afd14c70c4c2ba65e9d97ed91eb7694302"}
{"id": "train_035037", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "9bc5c244368db19e1f2d16fa65a6da150d8760ef"}
{"id": "train_035038", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "553f866c49cdefc0acb55fe4184c3b3da4808365"}
{"id": "train_035039", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "67d760708ea8936231e5788ecdb8aff4ec6daf0c"}
{"id": "train_035040", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "bbedb87efaaad718f73a4b234fa79165b0d4c422"}
{"id": "train_035041", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e5fe2caea7aab2b98450602de835331a7c6a6c05"}
{"id": "train_035042", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d357f3b1cd40fa1ebfc93e3434393645391f5797"}
{"id": "train_035043", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "468ac2739014d4a4d943c8c6dc8ab49b511e0714"}
{"id": "train_035044", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "3bbc858d9a7079141ec4daef1d89f808f626bc0a"}
{"id": "train_035045", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "052027b9168f83005ef5a76cfdfcec76c5f894bf"}
{"id": "train_035046", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8292b61a2cc9e7b7c85f82b7f7647b3c4fc76ec4"}
{"id": "train_035047", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f9c926cd4882c00c67ba97962bfb6a6d24a0b920"}
{"id": "train_035048", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "386fc4c56ed71194a00ed2676767a289361bdb4b"}
{"id": "train_035049", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "3c418f65609641d9b2811eb17ae789c4f5d04728"}
{"id": "train_035050", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "9a561e4412cd8548cee9394248e1a86fec2e451a"}
{"id": "train_035051", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "22e5bbf4414256f6b4ea6e838356809b5a936810"}
{"id": "train_035052", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "09bbafba614366230e5e2eb036ec6bf175e2f320"}
{"id": "train_035053", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "50616df38e5e36c02dbb4aa90e285f5691d31a52"}
{"id": "train_035054", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "39afc514d566f836637ac1ea20540b47f5384109"}
{"id": "train_035055", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "02e6b85dfdacb63e6a75207d1c3d63f1c93409f5"}
{"id": "train_035056", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["research", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "aa9bed336083a71cd5a8d79fedbcf5cc8ecd22be"}
{"id": "train_035057", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a872f7b2c2091f1c226602bd0383f7364ee9616a"}
{"id": "train_035058", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "fcaf289e61b88d72cb6c5e786fa34425f662bdbe"}
{"id": "train_035059", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "8dbc74bf8e34b3aebae0a93a5c3dddbcbc7f9ac0"}
{"id": "train_035060", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "9fc9894e79c0da4fe966d43120ded9aca8c9b84c"}
{"id": "train_035061", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "63bb526bed5c6bd91bc9b20b0d5d5b0175849f63"}
{"id": "train_035062", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "eb6e9340c1b452b4a0bb96068ed4bca5577c6c77"}
{"id": "train_035063", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "e37352725fa2a61d2b16d233ca164067b1602535"}
{"id": "train_035064", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "48e6b4fb73bfe31ec66531fd7a390ff642495726"}
{"id": "train_035065", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "711f4799ff1519d217c9be0062ef1124ac6acc0f"}
{"id": "train_035066", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "d01e53a6ee54ef64619b9e66cba9990120887fe8"}
{"id": "train_035067", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "5d7f06c6b34e5ec4943b640ff157fc6ce99fc77f"}
{"id": "train_035068", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "208db9e38d0d193edf045a86d25171369a44d56c"}
{"id": "train_035069", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b056c3de31a55e804e32a9ef1d62c966f64fcba1"}
{"id": "train_035070", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a77eec75e3e385301796dc2a75e143b84edc64c5"}
{"id": "train_035071", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e606cca094644f49ca7767542f30303147a62c69"}
{"id": "train_035072", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a00d2db7785982faa680e4101acbdfdddf107ef8"}
{"id": "train_035073", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "52c5f2273c5c6b3c29857afce1e964325f6ec6b2"}
{"id": "train_035074", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "11b2eaf000ae19dd385abd8b8d73134d485df5d0"}
{"id": "train_035075", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "b682703ac19ca74c9009eac8a2e05731d418b111"}
{"id": "train_035076", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "d225b742e01776cd9e97b49b4a51f42e0061f357"}
{"id": "train_035077", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "512bb7e2833ba898ac348fdf16a8a5c7bc89946a"}
{"id": "train_035078", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "54080d28eb759e7779e8f1e4aab2dbf5703e9616"}
{"id": "train_035079", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "e5737d83de231b9b38aa8e776300eba5fb16002d"}
{"id": "train_035080", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "6a44646789349f01ebce65abc15eea70cdde27dd"}
{"id": "train_035081", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6faae427fa5d212d9dde900bbd4826e48d02ece4"}
{"id": "train_035082", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "766bfcf46a68307ce9258e544dbbafeff7b58418"}
{"id": "train_035083", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ef79b89d9ee03b218457d5504fc44dcc974514a3"}
{"id": "train_035084", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e725c96b376c7af6956d9679968866f4f961f026"}
{"id": "train_035085", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "a5f05ab1f421cba1d5fa60b07bd2901180763af7"}
{"id": "train_035086", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "fe177e8f1b523044b6874e1d5c70039d29c6e891"}
{"id": "train_035087", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c264fa2091438a81262eedd27aea7fcc49b70055"}
{"id": "train_035088", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f542dafbc6e501449a94c81d16688d840a175bf4"}
{"id": "train_035089", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f507938f1a00e277a79cae7aa448bfa3f92f2ae6"}
{"id": "train_035090", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "1e0b24fe6543b3b17b6756314a93e0c0431755e0"}
{"id": "train_035091", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6018b4a569874fdab152796f02e727efeb4ead2b"}
{"id": "train_035092", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "42dbeb40c59704f361510c6f6bd75d73cef00d0c"}
{"id": "train_035093", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4ced41f7f79c7a74f8ff456125972fe6d2a30a80"}
{"id": "train_035094", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "e0a52c61a0b7261a746368b732fca9062585c27d"}
{"id": "train_035095", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c5adb2b00c30fc675ed93882504f13a9bf83ecc9"}
{"id": "train_035096", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "101231df9d3a763bc8aae35115d1e3fc6c3be120"}
{"id": "train_035097", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "766bfcf46a68307ce9258e544dbbafeff7b58418"}
{"id": "train_035098", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "420737bb767358fb741869f8563ab176cd4b87de"}
{"id": "train_035099", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "aadaf342210c5f5289f6f5cfece18660d1f0cf60"}
{"id": "train_035100", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "946c5268a8ca9abaef3b914694867b6cf54a16db"}
{"id": "train_035101", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c8d65a9384a6238e4de2d5582b7f6aa6cf182781"}
{"id": "train_035102", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["research", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "c951e5c2ecb0e21f751fd9bb21331ab52eaf61c9"}
{"id": "train_035103", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "24747a32fb3b0b5810cd5fb4eb38ca81c8438a4a"}
{"id": "train_035104", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "3ee04c0a914b962ef66485db9f1de5f7178ad63b"}
{"id": "train_035105", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f700eddc174448d3d61bc8734d5a38d409f9c6cd"}
{"id": "train_035106", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f42adcfc671a9cfe1b96e2820144f07daa403fc2"}
{"id": "train_035107", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "e97e6ab4d492a7de362f4a42474bb1893f4e8559"}
{"id": "train_035108", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "782096036a4a77baa025c786d1ef6301bdb54b38"}
{"id": "train_035109", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "b12f441e65d7d63c21f5f3e5c887fc68f0112aad"}
{"id": "train_035110", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "23b7a34725d1addbf6510d0460a6d9f4fc500391"}
{"id": "train_035111", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "0317b8417a2e623cbeeea9512d5d44aa5bf52574"}
{"id": "train_035112", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "0a7433e4b5789ebdcd9a684538706890fe639816"}
{"id": "train_035113", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "bf54c5a88bda7485bf25359740107ce11d308478"}
{"id": "train_035114", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "446376d0428adaa62ec60f21f295e9e9c66d06bb"}
{"id": "train_035115", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5d5dba248cddb54394e1d5ed8291f0b94fb59abc"}
{"id": "train_035116", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "824ee22dc101327bf2ef05fe52ace6a3c1591af1"}
{"id": "train_035117", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "70e99adf582b30a8bbe0f36e7291270add81ab7e"}
{"id": "train_035118", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "9e769f1b4dd5af23f351e55c923dca9c8b53673a"}
{"id": "train_035119", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "20a4cca2daa874407ecebf065d3380f07406fbbc"}
{"id": "train_035120", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "49ad207012c3f6a922a903708b064e69abe32840"}
{"id": "train_035121", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5f60bd1f9b8ded5303eec07d0230d2372510b375"}
{"id": "train_035122", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f57dda6bf8e6d0394bbcd51879a3914d42060d6a"}
{"id": "train_035123", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "1cfdf247a49c7c50d0ef9592bad0e5f618d22f11"}
{"id": "train_035124", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7034a84e72389ee14cbc62bb870daa92c4c5c4ea"}
{"id": "train_035125", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "9f7c0ef414418fd0156f3c0dab53d0acf7d893e9"}
{"id": "train_035126", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "1030c0f689be5094937b5df0a31e247c45240e05"}
{"id": "train_035127", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "ece07dc8edb94cad198d6df36d8ab30f4b7f56f1"}
{"id": "train_035128", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e719eb1f3c3a0e62c042789b718a1c4386789a95"}
{"id": "train_035129", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "5d5c4f56528f591947c93289e967ae7288e54e05"}
{"id": "train_035130", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "67d1264110dc921dc18d0440ea96609c0a799acd"}
{"id": "train_035131", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "39fe795d006d9d96434fb6cb67f6bf3f7fbd3b96"}
{"id": "train_035132", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "97f206f9e9b3ca8946df475b741db79c829832d6"}
{"id": "train_035133", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d5bf13df25168717d95b1c2946900e8d5a49f889"}
{"id": "train_035134", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "8c8417891056429d3e05b7932880289c72e2012c"}
{"id": "train_035135", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f34e7fa68a57e7eb286b35519d7625b43212ef81"}
{"id": "train_035136", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "0bade7dc740a0926c1d3191f2081f9ded861ca81"}
{"id": "train_035137", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "7c4fec3d89ee584fbf021d6cc9ece1c92bdfda65"}
{"id": "train_035138", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "529a5b86e9fd76f738230b1c66e62557f806f0a5"}
{"id": "train_035139", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "5c6be5751c96ee0af8df7d5ab1a902c957f975ae"}
{"id": "train_035140", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "7f488ba2ad69da98c8b6fffb7f32e4590509f1ca"}
{"id": "train_035141", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c7e959ea1a6a5a0a4f0c7c243f98b765adc9886e"}
{"id": "train_035142", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "8f12157e17e3088c272bf8c213bde54ed3452bd1"}
{"id": "train_035143", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b88ad6ae525d1721e46b76230c1c087675af08e5"}
{"id": "train_035144", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "030a82835f3535a12d1e444c369db84da74c52f1"}
{"id": "train_035145", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "df514ccd4c0d3a528f576112a6dc4e8f3c9f1c6c"}
{"id": "train_035146", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6be3475790b646c71d3032d613266c56289d9aca"}
{"id": "train_035147", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "434358cfe2a4997e3e9648dd5020eaa44762004b"}
{"id": "train_035148", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f1d8a04b4541d9f5760a26306e4c3dd2627a1b06"}
{"id": "train_035149", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "44bf85e42479cda900745ea73a1c75276a9b4a84"}
{"id": "train_035150", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "d60cf066c8c12e84170cc9c829cad5981f093688"}
{"id": "train_035151", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6bf4d741d14b755355f6420e7cb33992187b5866"}
{"id": "train_035152", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4704d7c140840ae41bb45c5714cc17679e7627e1"}
{"id": "train_035153", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4012fb4a145746422fa4a20e1974b6999e444a45"}
{"id": "train_035154", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "8290af07c4dab78e1eed6406a3831e8f48859118"}
{"id": "train_035155", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "99f3b7ddecd75016c8bbe52c59e031b183864b25"}
{"id": "train_035156", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "1029e6b7b6015a72113d3e552368621f23a48c22"}
{"id": "train_035157", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0d3065334f0c18f4f5bb5170c43b76f41ae791ba"}
{"id": "train_035158", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7a99d1321956bf44557ef5750daccc7dbbb67ee1"}
{"id": "train_035159", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6abdb25ba8a0375b063cd3306780fc267855bc65"}
{"id": "train_035160", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "0a918c4438701717484a5c63418d4d999521ce1b"}
{"id": "train_035161", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "7e837902a8adefe498082c2b32d4a11c78232287"}
{"id": "train_035162", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d008029c3500353b1bcb09fdee1390deafc1c1b0"}
{"id": "train_035163", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "7498444fb770d460e70039b39c2ad391db23e90c"}
{"id": "train_035164", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "e588ef1ff8be9adeb6b3327ed76fdf2ad26a1977"}
{"id": "train_035165", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e6de9bab9dc4e4b704989968d867f2565e765974"}
{"id": "train_035166", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "892413a08cf9768ab3da9c16bf297e33fc353a89"}
{"id": "train_035167", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "0224e4536da01371e6154e168fc717f896d90844"}
{"id": "train_035168", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e94e67705e8145881eedfa888fb70c5ead3c9441"}
{"id": "train_035169", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "e03beb3bc92b725f79d9e579da291181409f292a"}
{"id": "train_035170", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "8db1c37c9e0e095829fd41340674ec7f515a381b"}
{"id": "train_035171", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d4024632aff5ced9f0209f1785786898e1c1cc0b"}
{"id": "train_035172", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "e991407f386595ec4531957c9a93c1c084b3aaae"}
{"id": "train_035173", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "8b1e89d1eb4a57baf5a56903d58270a461a13348"}
{"id": "train_035174", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "45f421a2c0f4e4cb8f149c159c3d6236c2bd19a0"}
{"id": "train_035175", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "cf22f918839d38f4db79923f4a6f6dbc13d99681"}
{"id": "train_035176", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e2fe8a0916909b7510bcaa86aedd73da183c27dd"}
{"id": "train_035177", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "76dd58236d95e393a78d037f3fabf4898a8ba2df"}
{"id": "train_035178", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "075aa6ee201149592214955aafa9eb9e780489ee"}
{"id": "train_035179", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c07ca4f5c50797cf7d46b0abe389795fb7d79132"}
{"id": "train_035180", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "a05b737061843216aa474c9bb58703c5a19baf71"}
{"id": "train_035181", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "655b55c776b4b5d6c1fe58ac62d4442b46427336"}
{"id": "train_035182", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "cbc3236cc05247b12e20d155083dc3a23c28fe71"}
{"id": "train_035183", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "e5a956497da098b4b4d12d1181901c56e7ce31d8"}
{"id": "train_035184", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f6a5221d9ce41bb00653d5d99d37f4e6194e32e2"}
{"id": "train_035185", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3acb2df603e226a15768d0112eaa6908c798a4b2"}
{"id": "train_035186", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "08dbbd96862093a10b5cd34b7786adef4c47ade1"}
{"id": "train_035187", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5160a31db57211ea02af3e7c1f0249a1b2c3db1c"}
{"id": "train_035188", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "462bd0a213ba52ec1f16253f9ef9e52d62bcb3fb"}
{"id": "train_035189", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "27dfd2ab45d5f0ad233f564a675853eb33f080fc"}
{"id": "train_035190", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2a9568efa981f0ab927e0f2d766882e897f381fe"}
{"id": "train_035191", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "02edf614681928a84bcac469560158a9c6121db6"}
{"id": "train_035192", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "580e3764f58b2edfeed84d9399a5a142739289b5"}
{"id": "train_035193", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "ac99f685dba3e753b99d070cde0d0d83bc047df9"}
{"id": "train_035194", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "4508573d501354eafb9f3bdcd97af5561a2934b6"}
{"id": "train_035195", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "db5c4a4561ee49069d6cf8406eecb3895da357c2"}
{"id": "train_035196", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "8c44c3c65b753fb0eca24f7d6fbae0d8b66cd54d"}
{"id": "train_035197", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "f4861990ca274c58f0904bb0706d81402225c778"}
{"id": "train_035198", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "99ebf4cd96a94b29b0af2ab61e15b2e604cb92f0"}
{"id": "train_035199", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "b731a52997f62ffb43808082561a9eee726d7353"}
{"id": "train_035200", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "eb6e7685725710adc5be99e21abb70d33d041272"}
{"id": "train_035201", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "59a5f0a035a75990af00535dcf976da513eca136"}
{"id": "train_035202", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "bf782c38327a112de7e1b6abea404f90d84283fe"}
{"id": "train_035203", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "40bb2c8483fac63641110f70cc1a452006049819"}
{"id": "train_035204", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "410e5b62c71a500c301db04bd4dbb018a59efd93"}
{"id": "train_035205", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "b575d5a6127a4e920e67ab611e2b2a447f53a656"}
{"id": "train_035206", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "b35a2dcc23055919149a00002f8e7995f8fdce92"}
{"id": "train_035207", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "8410d6a036765298d7ff7d6e32feed8f22f6e0ea"}
{"id": "train_035208", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "676f79e1a6f6fe81808d28cbe9892b00ff0aa049"}
{"id": "train_035209", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "8dbd9a251137edfc333b6b0303f1fc884ff485f5"}
{"id": "train_035210", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "3be6b7b857d3b481fec2406b3346dd888392af16"}
{"id": "train_035211", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "1a83a3315d1ef7e7c24500b5fab8634306a72ec0"}
{"id": "train_035212", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b4deee2d0aa02a23442a24687ee00c06e612b784"}
{"id": "train_035213", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7af3afc51cae0b4ffb1dd194b6879bfef118cc18"}
{"id": "train_035214", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "8daae237e705103a94607f818d2189baacf8f3fa"}
{"id": "train_035215", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "52fe928a3141a36cf7d285de3c4bffe1c9fe2ba6"}
{"id": "train_035216", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "baeba7d5bfbe388393beced0ffaa00c3b3cd61e6"}
{"id": "train_035217", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e03beb3bc92b725f79d9e579da291181409f292a"}
{"id": "train_035218", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "4e7c54fee0966595aa9839cf149035c1d0ea255c"}
{"id": "train_035219", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "bf6eb5b4367f51f3bb020d222b75a17c1d43e028"}
{"id": "train_035220", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "60ba8ea401d0bf0e06c134e0a035890c102b97f5"}
{"id": "train_035221", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "47696c92db6a8ab815e735f06d7c73a8fb257a10"}
{"id": "train_035222", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "adc4b6f7836422b53e9808fde6c6b8bd2f9aee80"}
{"id": "train_035223", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e57ca0689b8494c2120d12e529c07a2595e541ef"}
{"id": "train_035224", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "bff12f92dfecc73eac2f0af1eccb513b4fba0118"}
{"id": "train_035225", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "8a914c85113bf4be34d27acfc4970eafe05e0df9"}
{"id": "train_035226", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "5049860b1d7aa1795d0410315822b57adab9cab3"}
{"id": "train_035227", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "67c4d9ef419aa976a6ef53fb2a786df6a960f2c8"}
{"id": "train_035228", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c58f5474b9949f76457cd72981f86c4ba821c56a"}
{"id": "train_035229", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "38e24cb070c541edae37067db6d0b9c21c514839"}
{"id": "train_035230", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2bced2a5113eb24b34c16ee0b8c55bb17677cf36"}
{"id": "train_035231", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["performance", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d511452825bb7959e59f9e1ac76540a1a2b2c062"}
{"id": "train_035232", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "fbb900d777cde3df4c6b2810fe5d7d7868b766c6"}
{"id": "train_035233", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "3879d8298d4f85a727cd82b6c45218c02e4a5e0c"}
{"id": "train_035234", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "0ad8eac2eb0c73913662f9a039bd6c3221a8b37b"}
{"id": "train_035235", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "37a3b88713b99d1119643d6147c501dbea487061"}
{"id": "train_035236", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "928936c073ea59dff16afa0d1f3a092ac96d89ee"}
{"id": "train_035237", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "5f63f94a926f59e3f4334ee7294383445e4085a6"}
{"id": "train_035238", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "7bd9238c07dbfad8686bfdba6d0b02e97f49775a"}
{"id": "train_035239", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2490dbef629598c6732995e30a1e1f48f2b25bca"}
{"id": "train_035240", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f17c5ba34e019682daf0a4bdfd5185e66e5f447d"}
{"id": "train_035241", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["evaluation", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "d7981bd6915be2a969f636cd75d9ec5459a1d7c1"}
{"id": "train_035242", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "6d41975d67ac474c99f84dfb79de943221a256d3"}
{"id": "train_035243", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "06ec99375af12dca663a74182b2dfa731d7cbb4a"}
{"id": "train_035244", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "5210f23153360b52e84176590f9652d9bfd8efdb"}
{"id": "train_035245", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "cc5baa7c3fed2a159517cd5acaa8e76e9cb40a26"}
{"id": "train_035246", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2fc5a81faf30fbdd24c1436ac9e00e4a98199712"}
{"id": "train_035247", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "cd598faa60dda991a6ef7c145763f7966cb654d1"}
{"id": "train_035248", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "521089336fc689af60aea7d830939f6b0257bdc0"}
{"id": "train_035249", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2ec6001b62c95241772c67747bd9f4b72e791bfb"}
{"id": "train_035250", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "85bd455faa6c83df067ca386fb1b0e8b125f57b1"}
{"id": "train_035251", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "f34d755dc328d05fe63a889b86f38dadc8fd56b6"}
{"id": "train_035252", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "8b47ea968a573c9649d3b02145b8bfe5c5de3446"}
{"id": "train_035253", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f5575fc6818ac86d66168760f255454f5f163fcc"}
{"id": "train_035254", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "b1491183be337c3552b630d53c4fb7b4e197b88b"}
{"id": "train_035255", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "b0b6a3b7d4ca88a23616b1a4ab385f7dc2fba460"}
{"id": "train_035256", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3c068f4dd88cd7e394cec2228f4e3cb2d7c5caec"}
{"id": "train_035257", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "174120f7c063836b65ccf5545521a7d5c924bc3d"}
{"id": "train_035258", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c665bb4a0ac8e44deb3e315c4f19c700dfdc5863"}
{"id": "train_035259", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "22568f77bd04a421ec0433bcbccd89499ac215a2"}
{"id": "train_035260", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "6ca95b1f34a86b3b560ef4f766f31b1640036f79"}
{"id": "train_035261", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4676b112adf58de4ce073ab8f47c71f7245bae7e"}
{"id": "train_035262", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "36d8f2debd6c64b50b65785605a33f48c9ee925b"}
{"id": "train_035263", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "dfa9d8b937fadecf2ccd38b4ba262b092ad8f7a4"}
{"id": "train_035264", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "60f3a997e27c0c00164e8b5efcdedb70ba2bc34a"}
{"id": "train_035265", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "a08ebaeb7a6a6e3acfd461a92f9de6647d99371b"}
{"id": "train_035266", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0369b9b0ded9e9989aea13787466e552326eae74"}
{"id": "train_035267", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b2caa2662a60d3cc55580b789644340519743a4d"}
{"id": "train_035268", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "64a655ffbf298d843456b63f7bf1c3216d057de9"}
{"id": "train_035269", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7700aea648a78780f5b1c78bcbc75931401e0349"}
{"id": "train_035270", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "01f0ede8641d7d45d31b57ad26694d5fea03b732"}
{"id": "train_035271", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "ca9e2c24fdaf10cef8e76210c497fab7f05a6c8e"}
{"id": "train_035272", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "13d231473267471549db14b379d7490080ee567c"}
{"id": "train_035273", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0ecae4fc4bf3b38fe199406f6b20f96c386d8258"}
{"id": "train_035274", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "200f3bcd91125f116d3d4337d2963f1e4ca6c8a9"}
{"id": "train_035275", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "0420c7b3435be9792ae43c05748ebe61ed0a70b1"}
{"id": "train_035276", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["agent", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "c82a066d67c6a54d205ec7bd188ed4768aee2d62"}
{"id": "train_035277", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "030a82835f3535a12d1e444c369db84da74c52f1"}
{"id": "train_035278", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "2e3354489ee22720deb7638a417931196aa32b59"}
{"id": "train_035279", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "4d3ce87e1bafe6a56010e5119349bc8a3051cec8"}
{"id": "train_035280", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "4ced41f7f79c7a74f8ff456125972fe6d2a30a80"}
{"id": "train_035281", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d7ff28db8cc5ebc0aa0285e4795eb8d7d3b0beed"}
{"id": "train_035282", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "cb139e31f47786044570a7b6da17893d5997d743"}
{"id": "train_035283", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "3973cd4996bbd084f427023e47b30230b52f7846"}
{"id": "train_035284", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "d18e9bb89067431922a19a4e477024330cbb9fc2"}
{"id": "train_035285", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "23a004a4bae03966bcec673806959b305d5dd5d6"}
{"id": "train_035286", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "cdaddb9931dc6581a22fbf1267be3099d1224630"}
{"id": "train_035287", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "02c976264e22c53c1eb5c21a85ef70ef3fbc4b65"}
{"id": "train_035288", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "9877c5dbeb556c1badd34f5b405f2688b21436b9"}
{"id": "train_035289", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "10e314b505ccd6358a835bb5f24afdae81a620cb"}
{"id": "train_035290", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "c664fd32416037e33a135585cb20ad57b525ccae"}
{"id": "train_035291", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "916f8895aa83c5ebcfed036411ddc8af76ccfe20"}
{"id": "train_035292", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "233395ec8bccaad27797c62e9c11eef4f42615ae"}
{"id": "train_035293", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e660f230d924ed66cc0bd072688f5b32a351aaaa"}
{"id": "train_035294", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c7313c9ab967be021190407be8995781cdb20a28"}
{"id": "train_035295", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "79d98ad671a395134bcc57496df2502f53e19733"}
{"id": "train_035296", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e49d4971c85cde2b4aee4d4483c8ed538457b436"}
{"id": "train_035297", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "aeed382ab75e377d328392feb79db4e2a0c72a71"}
{"id": "train_035298", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d9cc9584947052604b76ea47cf862b94752e1adf"}
{"id": "train_035299", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f2a687cf4f327673eede6e7c7c2e3551b44d928e"}
{"id": "train_035300", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d8a2b82c1a4e1754d8bef63dbfba127822cb168d"}
{"id": "train_035301", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "edebe45e98f68dbae8031c13e94521a3ff0adb99"}
{"id": "train_035302", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b364b4cfbff735f923d7fdab7c2315141728c20a"}
{"id": "train_035303", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "263277ddd83eb2ff96444c7d50ade9a8be5f9779"}
{"id": "train_035304", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "7f791ac4684c1092d0fc6365c79d0f4d97ec6de8"}
{"id": "train_035305", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "097f1b175cc7aac816120becbde62f70f8909fa4"}
{"id": "train_035306", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "fe7927178d428fb3ae33dccbab48b1b6fab582ba"}
{"id": "train_035307", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6c415f1855a08a03391579d4b904c4fb5a3d51fe"}
{"id": "train_035308", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6718d222a22ad691de275c8687a5feeb126bc9ea"}
{"id": "train_035309", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d6ef3a50c67895e0e191b72bd840f2255c8a9cbb"}
{"id": "train_035310", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "9fa8d32500e1a8d5d12cfe21a5ece5b17d2624fa"}
{"id": "train_035311", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "957cbc3e7158e9f6f3f4b1dd35970996d0849831"}
{"id": "train_035312", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "ec86fbc9541c9d32fb80f8ca2ed28b54650c28bb"}
{"id": "train_035313", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "fd5db821d3d08e6df095a6ad92685ca99ac6779f"}
{"id": "train_035314", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "169bda21f9bd1f80480dc0e226ffb5b6791265f8"}
{"id": "train_035315", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "2ac64879e7be3bc380b6f2dad84b63146267aad7"}
{"id": "train_035316", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "46c537a9743c2084abd2d15300a965a33e21375e"}
{"id": "train_035317", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "fa947277d0603b74a1c127a5598d674882ffe7e7"}
{"id": "train_035318", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "93d21933b237640a504763aa08c67d3839dbe3ed"}
{"id": "train_035319", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f7cb99471f96c032ec4b6edbad4e222c5a2c5e52"}
{"id": "train_035320", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "abe1a0422853e92c5a8a6ac3a93c88f290e3732e"}
{"id": "train_035321", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4aeea3fbe48f37bee20aeeb801fa6a6690f10357"}
{"id": "train_035322", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "13b10e706d7cf5a7c6b9ab3956abc28ec34f5e7c"}
{"id": "train_035323", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f60aaef0034a3e284d62b44963f5be2f2385d748"}
{"id": "train_035324", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "8b85157d14bc5884a82789738187020b7b90f906"}
{"id": "train_035325", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "5bfca95102c26c9aaeb8e38d5d12246031412cb0"}
{"id": "train_035326", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "7783a5fcbbdef49f62f65cf7d69113d9317d815d"}
{"id": "train_035327", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a588b072b0f7cef8522adfd8a3190460e17010a0"}
{"id": "train_035328", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "8f1e62aff4aa092abb7110df3896e0a52bd6306c"}
{"id": "train_035329", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "6c2904f9d8d548822f78ae61f9eee9b767c1021e"}
{"id": "train_035330", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a7de658a26a4d941cfbbc7a62532ff5b5da5023a"}
{"id": "train_035331", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "6bf9d9d691d34898a2baf93db837ef644bbcf355"}
{"id": "train_035332", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f9316a172f774c206d5bc1fd57914f758c49176e"}
{"id": "train_035333", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5b042f5300b5a8cad9c832b8f54750e7b7dcc326"}
{"id": "train_035334", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "5ddf61c2795cb0b8d548a70079f25304758b64ba"}
{"id": "train_035335", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "278081b53c4fe663dc6360a6f962d8fbb217e911"}
{"id": "train_035336", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "abc5f2b11ca7e71cd6a71bfdd8d0af57c814dd9a"}
{"id": "train_035337", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8ef6c5bf727738587042669941ad5a41e06f5abc"}
{"id": "train_035338", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "da4023c189993160e7907b7733e25d428dacf037"}
{"id": "train_035339", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "2893fce8b76829f8d805f56ef86f54dfb83cda2e"}
{"id": "train_035340", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "0e735b74e01b693696dd34327a5163edea16d56e"}
{"id": "train_035341", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "9590546b4faa155b39e99d2ee2764cc4b9b9c322"}
{"id": "train_035342", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "b506e0484d887111d621d67223a802302ef254db"}
{"id": "train_035343", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a659e7e923c4f470c12f510e51c3949c3d64d07f"}
{"id": "train_035344", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "da66f6c91b274a0a491a189f4fe7c03a94b85d40"}
{"id": "train_035345", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d17d0b5732041073df5a9e5ff9d091ae2661ece4"}
{"id": "train_035346", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "82787b20d47a874d7f27caab355785e7c87b0d65"}
{"id": "train_035347", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f649df02df0bebadc33129156fbeba03b75f88c5"}
{"id": "train_035348", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "6bae441cc17740590b119c80623200895232d517"}
{"id": "train_035349", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "517a43b64643d1a39608715e99ae171119ffa111"}
{"id": "train_035350", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "851833a4894c07e329fad289ed3b3568dc200e47"}
{"id": "train_035351", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "424a382501a7f865f14388942b524865a225e497"}
{"id": "train_035352", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "9cad8d2db7a96cf886ede64355fd074d612b80db"}
{"id": "train_035353", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "e6a90ff22ecffa4a0e30e31714ec2b2317e4308e"}
{"id": "train_035354", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "52835234c5e6d800a442234bb3730360d97b02e8"}
{"id": "train_035355", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "eca07127a2934e179f6b9ba212293eeef5ac75aa"}
{"id": "train_035356", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e3e61ec2c22d4885f70166cf3a6b18383d1a92b8"}
{"id": "train_035357", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "1540d558ffcd7a1b6df058864ec75a7c354f7d14"}
{"id": "train_035358", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "8ec0d6791488fe2d22a3bb2b79d9986d3a46e3f2"}
{"id": "train_035359", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "fcb993d46e724c83d5df613ca81c912225043ab8"}
{"id": "train_035360", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6a544906a747ddb6c7ef249710bbf6ba51d62658"}
{"id": "train_035361", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0007bdcd575c43806e52d8cedb568a7ae32d6ef4"}
{"id": "train_035362", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7689f86a3df1d28f7237955a68208eeffefc0840"}
{"id": "train_035363", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "9291d0677ea8111dc7ccbf0456e495f48f52171b"}
{"id": "train_035364", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "1ef0506e100603eca5ccf97caed621512783a8e5"}
{"id": "train_035365", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "fdb6ec7c9a84c46f01831e20d20bef81dc769d9b"}
{"id": "train_035366", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "dda6e81e669c26180c85b0f2204fc2362e87b938"}
{"id": "train_035367", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "e3932c23ce0057bfe3068b4b45c5699febdec06b"}
{"id": "train_035368", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "b3f99047dfd90a0d5db4ede4910e241d96dda28e"}
{"id": "train_035369", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "5093fe04ae6ed227f7dcf371cb12b222e3346444"}
{"id": "train_035370", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e299ec84837330835f1af5ba3e69a070fb2ea6c3"}
{"id": "train_035371", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "175047e7e6bea86316327d32de582d45cc11f976"}
{"id": "train_035372", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "07b918e94e6c360db3ebb7e36e8ece2ac406427b"}
{"id": "train_035373", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f5dbb675c062aae94045c852a386ffd55b78c3cc"}
{"id": "train_035374", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c8e47811851385fd42391734077130169b1dd373"}
{"id": "train_035375", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "0cb8e73b6d8cd2db22474e3e890fb26f19bc585d"}
{"id": "train_035376", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "1dc491f79f71c52c0681b62ab2cf890175822521"}
{"id": "train_035377", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "5d2d3bd14e5e24f89e46a0763f302010ec96ace6"}
{"id": "train_035378", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f9bc3203723a0bcf476051ecb955e9591ee18579"}
{"id": "train_035379", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "2791f665838fae460710e97cc0fd6afb2974360f"}
{"id": "train_035380", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "630d8f949160adbfbb6ef4e9861bd8d47f210986"}
{"id": "train_035381", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "eb1166b1f7826e119a1b4d47e744b10df3d6472f"}
{"id": "train_035382", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "ac2b819dc2a4e9efa5196333d0ff61d435bc3961"}
{"id": "train_035383", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "38b0a278cf2fefb4c4b12c77ff126644d0c1e217"}
{"id": "train_035384", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "1a3f6e91e13df101de5b6b8da1cee8b43f2b9865"}
{"id": "train_035385", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7f92d07158011918f37e8e4a8a250e3463e9a56f"}
{"id": "train_035386", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "96755d0cc1875c963c583cbd88ae6102994c9e8a"}
{"id": "train_035387", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d87926c7c2f09a9f920e9a0a443d5906bfdfaa29"}
{"id": "train_035388", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "78071b3ab6b92be5aa7f9d8a1e7ae2b05bfd4256"}
{"id": "train_035389", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "eec884bc458f0dd08436b6352c21a5727708da2d"}
{"id": "train_035390", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "2c96f4b713073a50453eb16b5aa1b61b337c7097"}
{"id": "train_035391", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "77ced9a516746bbb5ead0bc45e66a347f2525df7"}
{"id": "train_035392", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "54e1d375bd623d70f90d432e84089926ce0bd6f5"}
{"id": "train_035393", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ea335da809db4488cf67dd5f93f5ea4bb66def04"}
{"id": "train_035394", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a3ef192a82f1f87fc4d113d8350623eaf4e8490d"}
{"id": "train_035395", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "8d9f042647d61bb6b9de786ea2afb9c8bcf2104a"}
{"id": "train_035396", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "9507e255e737133cd6cbd5baba07ecf9f587d23e"}
{"id": "train_035397", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "85558d4201e13e92ddd44089053da4cdf91d91dd"}
{"id": "train_035398", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "2bc3592679ad57b1ed3a63ad063961ff761696b3"}
{"id": "train_035399", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "d9cc9584947052604b76ea47cf862b94752e1adf"}
{"id": "train_035400", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["research", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "3a4731de0a9cd47b79151c79f1a64443a3428c72"}
{"id": "train_035401", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b1452502c9144a138b98fc85d355ae30da019a7f"}
{"id": "train_035402", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e259007daaeff209655608019e1727b7ceca2de8"}
{"id": "train_035403", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "e70f9cd4bd300a4a50f84292c8e78c96769bb4e7"}
{"id": "train_035404", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "01b888c77840d2684943835b83fee7a23b64e88b"}
{"id": "train_035405", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2637586b8386e4a44f52525770148f5cff0b14a5"}
{"id": "train_035406", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d3e2b46255c27a92e0443ec63df864a84001b97d"}
{"id": "train_035407", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d40a848a3005d98a0a8bc38e046454ede1ebe4dd"}
{"id": "train_035408", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4dcba614b274ab1cb1341ba3dbfa8dfc5a13e4fd"}
{"id": "train_035409", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6639da6417862da547e1a7e0c5f7f70b252371bb"}
{"id": "train_035410", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "1a883e1cbd77479c6ddd74463d9fc5c23a95aeec"}
{"id": "train_035411", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "b709d796aa5a5e039b51a790adbf38178fcc484d"}
{"id": "train_035412", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a3aa213b781326e3e8beef7869b06784a8094a1b"}
{"id": "train_035413", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["security", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "bd25d1bb807db30e2fde4c8c2e33a98906310155"}
{"id": "train_035414", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "5c4327c7fc2a811a6ace27580e4df3f6a653fea1"}
{"id": "train_035415", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a086921b13d892bfb5fd55885c7da5f6509278da"}
{"id": "train_035416", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "32c254d0b363ee69d4a554bf79149b84e2d555e4"}
{"id": "train_035417", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "72be3472126f9107b8b0a1d0df51bf07501f2ac6"}
{"id": "train_035418", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ca3a2760eec83643396909c0955c1ed30dd7a017"}
{"id": "train_035419", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "10cf5f8fa3e3645893f7bf88cbdce8944c909a9a"}
{"id": "train_035420", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "24aede1f79a65dc13eae981609843508ab076f94"}
{"id": "train_035421", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "371f3288855f70b212cc416068717f95ba3ba970"}
{"id": "train_035422", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "ee5deb8649614fa86cac03bda12ef2ce829e4dde"}
{"id": "train_035423", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6550118b434804804839c902c6af7900863a0566"}
{"id": "train_035424", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "8b47ea968a573c9649d3b02145b8bfe5c5de3446"}
{"id": "train_035425", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "296ae91328a36484b3ac1363e3ea2926872cd483"}
{"id": "train_035426", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "6add7d9277d50b4836a76d398264545546c599a9"}
{"id": "train_035427", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "35ad17b7911bb526d39267c17a1bfb8170450dc7"}
{"id": "train_035428", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "f8eb40a2a05c85395626382ceafa5dd3791861dd"}
{"id": "train_035429", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0d6f9765b53ab90cf281a231abc6122723585834"}
{"id": "train_035430", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c07d43d5e3d7eb1d288377b13af776dab2478cd0"}
{"id": "train_035431", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "3505aa8c8cd524072bce62d9134eafa4f9fea911"}
{"id": "train_035432", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f276ddb73191166e373ce24186a45b2576f6655f"}
{"id": "train_035433", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c8e8462bbbc5ca142e6ed6a99be285a59ab3b0bb"}
{"id": "train_035434", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f3f029f1d487cfb8842bc7d4dca60d2f40fc6efb"}
{"id": "train_035435", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "3396424c0a3f4b2d59d81e20ebca149f5aa87281"}
{"id": "train_035436", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "fb645a70c140e21e5bb772aedaf8db715652ff20"}
{"id": "train_035437", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "fb413e3e3e5984a9a39d4dff5d7039959e2be90c"}
{"id": "train_035438", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "2eb969197e85e3fded3c268df129b376b4e3bb99"}
{"id": "train_035439", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "9590546b4faa155b39e99d2ee2764cc4b9b9c322"}
{"id": "train_035440", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ac23099285c1eeb8d2369ebddd8eb39d522f632b"}
{"id": "train_035441", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "26191b38278f6b9f3cd619ab335e50e4d25109df"}
{"id": "train_035442", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7bd72a2d88e96d34748f85146836cbe0aabc8338"}
{"id": "train_035443", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "04cf721dba3a53c99f5dfa1226faa1b3194bb324"}
{"id": "train_035444", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "38225f08732a0d852d456643d533ff238207363b"}
{"id": "train_035445", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "4c76c2c55d95195415a88645541b25c23fc283de"}
{"id": "train_035446", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e222969d95e6a028e2fcacc3998ffc19873d1b93"}
{"id": "train_035447", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "188e46e0fdbeff045c6841f7c80cb3b70662f113"}
{"id": "train_035448", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "0048dcbaa676c76d514fab1a4d1a32ec4da9bf3c"}
{"id": "train_035449", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "55ac8e868de881c5f4e444008b58ca2a2c053aa8"}
{"id": "train_035450", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "91435b273192bdfbdff5b07e6a4082f3bb3927c3"}
{"id": "train_035451", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "fddbc3c4e3874b8c074e2270766eccaf326b2abc"}
{"id": "train_035452", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "d5655a59f9665ef96adc7d92c48e0bf6e739eeb1"}
{"id": "train_035453", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "5451324b514c92101126595dafac1c7ba50ef8fe"}
{"id": "train_035454", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2893fce8b76829f8d805f56ef86f54dfb83cda2e"}
{"id": "train_035455", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "43867e2b470d5683e74ecfd0e66456d338c1ae16"}
{"id": "train_035456", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "6e46ee81bba69b2481f008661b00c1c1e4675fb6"}
{"id": "train_035457", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "14ea3b4409db8ab9f6c57f1aeafa5fafb2bd9af3"}
{"id": "train_035458", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "29aa510ef40a10ffdc91f65faf20dad16e6b8f5c"}
{"id": "train_035459", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "3d5290e640f4cc2eebff7dd6a7069aa9a7174df2"}
{"id": "train_035460", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d9c55929cd72a4dd9bf45bf71a012ba2a06c6ec0"}
{"id": "train_035461", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "c1de3104cb9fa993b41e1fc5719587ea53eb409c"}
{"id": "train_035462", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "8e0273ec161d7277b15db018b88aefa333297a20"}
{"id": "train_035463", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c8db01edbf25496ad99a842357e7e56c46ebc5d3"}
{"id": "train_035464", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "b0b61ae8367d7fa650ff5106c5a9f1f6d38a4555"}
{"id": "train_035465", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "e22d0ee9ec118dbabfc1881aa2e9a5550aa7b791"}
{"id": "train_035466", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d8713c843de21311b84056c453c808490f4f742c"}
{"id": "train_035467", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "c868ad620e776f57dc0d9c6f4e9ee31ae79215c2"}
{"id": "train_035468", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "95f335dc0c41bd69be4da9cdba767bab76c9c6dd"}
{"id": "train_035469", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "fa78ff8da5f99d037ed78cfcfdb10b01ea6eb783"}
{"id": "train_035470", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0c933b349417011c3d5824e5cf9633c06c352ebc"}
{"id": "train_035471", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "2c7f6aa283fd2d1000ff913e300196ec0c6898d0"}
{"id": "train_035472", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e62116db586d92004e2aa9882ae1c16755df9c1c"}
{"id": "train_035473", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "9ca262315964e1bb5d02b21836c5054a335bce58"}
{"id": "train_035474", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ac59ff758f563480600e2d00879c89367ed5e5d6"}
{"id": "train_035475", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "769dbe9fee6280b755525d8e0a9dc6eb28385dbc"}
{"id": "train_035476", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "da2b0d59545bfb558d47da91212753fb6589a17e"}
{"id": "train_035477", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0c4b164cf4996d49a7ea6b1d2fbde6e20ba18c4c"}
{"id": "train_035478", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4c001b963af9ecde9d36065917db39e0ef7e5375"}
{"id": "train_035479", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "13c35c91526621cc0b97c4a8c79ab7a81b4592cb"}
{"id": "train_035480", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4af3b5e3d91a154877c7b41c25b304fafe5a7121"}
{"id": "train_035481", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "e3f280d017919269114d12fd4c8ca59030e9c8fc"}
{"id": "train_035482", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "0a04f134bc99df20ae1517a4efe46c9f0e991950"}
{"id": "train_035483", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "8b2810cea2c34f5954a2811e127eca20b84c539e"}
{"id": "train_035484", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "b2a01d1cf5c9c0d4776d50464098e9919ee9bd58"}
{"id": "train_035485", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "b7f28158106a9db8a7c2b221ec1813f5ed9cde64"}
{"id": "train_035486", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "57f0697dd57be146127536516c915466ff52a611"}
{"id": "train_035487", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d1f29069b3fba5e06fd9627e97da3681492f460b"}
{"id": "train_035488", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "58ac95101aab8726c77bd68e4b0a3fe76a12fbe0"}
{"id": "train_035489", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "8e103843d5380983a087c5dc76efcf7c71b7b748"}
{"id": "train_035490", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b9150124c9f85120997d2ba649ca7bce904ca03b"}
{"id": "train_035491", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "6ea320407c76c707ebf020d46b99399520f25fcf"}
{"id": "train_035492", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "6e5c7aee82186c31c9d74e0abdc2870c0c623921"}
{"id": "train_035493", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "b8e6af3100f315e6077a155873d04287ccff21cf"}
{"id": "train_035494", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b02e54b39fa8a7e4046b3d4ef435ef6f6ca02187"}
{"id": "train_035495", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "76d08805dfc19d06439e6debd321fb842ac7101c"}
{"id": "train_035496", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "688c5aa2d7b927fde112358a809e4c3ccdd8ec8e"}
{"id": "train_035497", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "e128efaefd3d2957c9bc7287defcc4107f2cf387"}
{"id": "train_035498", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b95151e0d7d9a27f59863d4716fcaa60c7ab5b7c"}
{"id": "train_035499", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "130475008565040bb37219de91ddcde617ac0d03"}
{"id": "train_035500", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "ea54782a51569b610176734ace0ef4f526e850b2"}
{"id": "train_035501", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "a26d367c085e0c10a7cfe4dcf015282567b1ee21"}
{"id": "train_035502", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "508ed643f92c3c289d99f508d9611542e3e13ba3"}
{"id": "train_035503", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "28691fa73a02aa0cc4cf7fa0daed02d044af3659"}
{"id": "train_035504", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "1bf9c89191659fa90f835c6bf5ec73495cd5af1c"}
{"id": "train_035505", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "94ec4202f4f02ae5b45a15b65cdbc1b46cde2de4"}
{"id": "train_035506", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d5df7a9b5b0d724e0228668d43a48287ed106d23"}
{"id": "train_035507", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "017535ed54e36e4735cb6796f254fa0bc2359912"}
{"id": "train_035508", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "cdde60891cd6aa708572f6d1cb1320e6244086ac"}
{"id": "train_035509", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3c319f3781fb1e05565ca16cb99eb3c01edbdc89"}
{"id": "train_035510", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6c2c650b51fcb18419e25d2715bb4c415f28b8df"}
{"id": "train_035511", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "7f92eccf8f911441e21960955dac028e32d36610"}
{"id": "train_035512", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "3384987e2cf7dbbaecf2c38bb054d08a7567d5ec"}
{"id": "train_035513", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c01e2a88423d8c5860ee456e8e64d5bb98c6d3e9"}
{"id": "train_035514", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "941eec5767a3a9f2dea65594590424717ac3c276"}
{"id": "train_035515", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ac81d0ddda075c6662bad332dcfee5d3917229fe"}
{"id": "train_035516", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "9efd66dd3a33013642e19efba92a50fbad0eaee6"}
{"id": "train_035517", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "741559f10f38dcc96e18bd25336f0724357fea80"}
{"id": "train_035518", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "36fc18555d6145d91522acbf18935d5bb6db4db5"}
{"id": "train_035519", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "487ec22aab1af0795a8e6a2cf732c7c4b20f6adb"}
{"id": "train_035520", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "9c896f532384e5f56aaa5f032fc89f6a9b27cbcb"}
{"id": "train_035521", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "55245f5ee4ac93847b4591d346c3624a2ebea6f3"}
{"id": "train_035522", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "b89abe01ab60e71913d270cd268fea33f450e75b"}
{"id": "train_035523", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "641b04142ea40fab0976606175e70128e083a280"}
{"id": "train_035524", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "eaf9d23589b6bfd7af27925ba13882cfc70cfb42"}
{"id": "train_035525", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "7568f2217e87e3cacbb1a70e8b6409838641e3f7"}
{"id": "train_035526", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "21b57bae0dcdcc3a6062669eeccf49e3c933f07b"}
{"id": "train_035527", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "c2996ff23231a0abcb30598f0fd52d2a9251f731"}
{"id": "train_035528", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "a24306d1bc296502d20d442ebd17c4b3568a7644"}
{"id": "train_035529", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6aa0aa9a4c45d7e833d4e8ee905dafdb4924c573"}
{"id": "train_035530", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d224da9c218ee788e8f57181c122e65aadf0b388"}
{"id": "train_035531", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d621edfcc538bf92d395809e971618ca648474db"}
{"id": "train_035532", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "abff6c3343c8cbbae0b6b73e00b7cc9aa6749d34"}
{"id": "train_035533", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "ba1cfeb911de521e726a097ec22cc1784d46f916"}
{"id": "train_035534", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "e991407f386595ec4531957c9a93c1c084b3aaae"}
{"id": "train_035535", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "1fb321dd0fcea0135702ac732558e56ee67b844e"}
{"id": "train_035536", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "3a660bb992b8ae2ae9509ff0d208db6db01d37c3"}
{"id": "train_035537", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "feb14c31ff7538927ffd10f342a7d57f7bae9dd4"}
{"id": "train_035538", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "04ae405434bc1efd08adf401f29154943210457a"}
{"id": "train_035539", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "fb1e5c60d019dbcc9d6dc19cc4f3f5a0c66cce4f"}
{"id": "train_035540", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a5f466125aff5f3f85bf85fddd2241215963e79c"}
{"id": "train_035541", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "72ba03049e581aa64e4963c968e3ee4ab643763a"}
{"id": "train_035542", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f6423adcd70bb7f5406407f3e9e6cee39f3b929c"}
{"id": "train_035543", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "fd106ce0120241d47b7e8a8f4692081dca134851"}
{"id": "train_035544", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "568a818c88c1fc9a116acbc6c19a00dae5319461"}
{"id": "train_035545", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "70f5c75a9acded627fccc423136ba89a893fbcf8"}
{"id": "train_035546", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "84ff97fbbea57d7687982bbe2b202510475ebd32"}
{"id": "train_035547", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "97d9dca1088e97a7dcc7714de3f3048a0300f0d5"}
{"id": "train_035548", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "cc2acf8b8e7949dabaff5258fdd3394449834fb3"}
{"id": "train_035549", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a6e6bb42214aba0e44edfd532c420b686f2ba908"}
{"id": "train_035550", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "efe5ccec47722fc31ca520fd5a93ec392799e249"}
{"id": "train_035551", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "489335b0806d7dadfafe9ff7dc3eba0302c8f09e"}
{"id": "train_035552", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4c71a0ba6249a29a2767bb61ad4e46b6b4a7f82e"}
{"id": "train_035553", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "47e942fae6f6453b15995754161fe5a07fbf3757"}
{"id": "train_035554", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "c8e8462bbbc5ca142e6ed6a99be285a59ab3b0bb"}
{"id": "train_035555", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "dc45df23b49c682bd8d8b03bf3e37e08ff534653"}
{"id": "train_035556", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6f9cc61c4774176d9b450caba6952d97f204b953"}
{"id": "train_035557", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "4effa13ee32873c08999e3a0da18785e4600155a"}
{"id": "train_035558", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "7491d5629fab28dd11ce5dc902b95cc1f4209b2e"}
{"id": "train_035559", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "b89df570eb320fe17bc1aae04f89057168ff8654"}
{"id": "train_035560", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "a82661205c002dffcc5471ab384ab1369a83db21"}
{"id": "train_035561", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a9fbfd5be32368b7343444e840cb7ac2240f658d"}
{"id": "train_035562", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5b91297cd84ec7ecd71456ace2c326785f8f8f42"}
{"id": "train_035563", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "31c18fef10b18d3cb5838999ed873cd85ae5bd32"}
{"id": "train_035564", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "20e8766b94a34efac577dbfda7f8fd0aad80ebf9"}
{"id": "train_035565", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "53c3acdaedf49ab1548c7c224975b24b81f98a7f"}
{"id": "train_035566", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "58e68b4f04894799acb395f27c4455d11178b9ba"}
{"id": "train_035567", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "aaa059099ca34f470740dc63d5cc35a2faf30a14"}
{"id": "train_035568", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "1c3f084cc61964a166493008055726b7272cdd39"}
{"id": "train_035569", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "6625292e72bd00496e5f1e8dc510d92f0a860965"}
{"id": "train_035570", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2059e5c1317333502c960a79117e7ca8a46308aa"}
{"id": "train_035571", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "16368252d4a828d8853b04b31707a67071dd167d"}
{"id": "train_035572", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "8a1ac9c67a3597a49f3f4897106d55049c25abd5"}
{"id": "train_035573", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a8999b23e59d4b6fd3851b2ce22c126ae3a8c0d0"}
{"id": "train_035574", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "166ede37be43d45963727a43a85fb502dadf6f32"}
{"id": "train_035575", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "52108ebf81950fa488771a78a2603511d3a0de87"}
{"id": "train_035576", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "301ac523373da6c37e620328fb57003463f40cd4"}
{"id": "train_035577", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "9d179aac646a1d9fb5d7698b661a30b64d7c44f6"}
{"id": "train_035578", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c8ef9716bc17987f479304e168189560d9b1f49e"}
{"id": "train_035579", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "47093086290d2312d19ce02d9ea7cff864d8fe75"}
{"id": "train_035580", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "dbe5d769d043cc1c61dcb6fc3ec2feca88316083"}
{"id": "train_035581", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "baf817791e9c11944838e7672e93034071b10ff8"}
{"id": "train_035582", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "58a7a00d0e71e8a699e79fea564378280ee3db59"}
{"id": "train_035583", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "e6850e89a2cfdca98a009d2205bf583ceb6140ab"}
{"id": "train_035584", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "6efe83ad2deb99ec069a23136ebf6619413fba17"}
{"id": "train_035585", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "5365a1943febd54c3c3f34d3511543eb0fc515f2"}
{"id": "train_035586", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "54e51bdc5d8e9b6f2a40f68ea5d562935ffcf61f"}
{"id": "train_035587", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4b328a805ae172d61548311df313d5477d75b60d"}
{"id": "train_035588", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "76deee9e4283ed68910950c5b5b186101c591d18"}
{"id": "train_035589", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "83dea4ab8f0cf3e99cc216d41c9aea1d76d3c670"}
{"id": "train_035590", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "cbd612e6c6caf25940fb3baa90b9f2af24955d35"}
{"id": "train_035591", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "87691735f9a991eef8493e107c2384f9a7dfd999"}
{"id": "train_035592", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b7ef183b50381f5085d45d921762e8a695b4f1f3"}
{"id": "train_035593", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b626b048ffcefab197a320479d2c4a38146386ca"}
{"id": "train_035594", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "1c597cdb234ac747232cf03391bbf9e18f7160e7"}
{"id": "train_035595", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f368c06eea87c0cc99a40bbce8d06cdb8c6ac7e2"}
{"id": "train_035596", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "7b6f1a432037227850cced06bdfe97b28dbd24b7"}
{"id": "train_035597", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c40cc7044dca3166976cc10b1de9d992ac38e4aa"}
{"id": "train_035598", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "cc5baa7c3fed2a159517cd5acaa8e76e9cb40a26"}
{"id": "train_035599", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7df38f78c8bf6e90ae85375554fcc359659f18ca"}
{"id": "train_035600", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "3ffc8cbfbd9c6c0afd74764d6b49459b7655cb9f"}
{"id": "train_035601", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c42b10818464b0bb07158f14e167b002b5e29c64"}
{"id": "train_035602", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "97e018d2570f56ffdb8128ec705687afed0d02e5"}
{"id": "train_035603", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "1e6405bca8aadcf0d23bfab7f6a5567604004e8a"}
{"id": "train_035604", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3bad62da9c3792a40648b0e1f7158f392e1f388e"}
{"id": "train_035605", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "7869863c1459f8ad8477ded6ef7f1bd7bf8d6fef"}
{"id": "train_035606", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "dcf0c03b821b56bbe03c82349805686f2114dda2"}
{"id": "train_035607", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4f23aa62fe6420adddae9f95c8531d68a738e0c2"}
{"id": "train_035608", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "2e5d14f02a12070c9486d43e4a1b9942b71267ff"}
{"id": "train_035609", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "564d1850c3db42ded9e66110f0cd9e9c3acfa351"}
{"id": "train_035610", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b6757a2b71785ef68bff46ee6dd6132a3bfa2a2d"}
{"id": "train_035611", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f7eafae2c5f202bac64ea29e5de96b5fa52f7b9b"}
{"id": "train_035612", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8eb5f4e48f26dc6d474ab3edffe68ff13ddbb2ed"}
{"id": "train_035613", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ecd3dff5b82684eee6f1855b80dd3697074bd2a8"}
{"id": "train_035614", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d60bc1ea948006d53dbfba99ed1d7ac5002c02a7"}
{"id": "train_035615", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "7434996713d7ecc7f491b498bbbbda5e0a5a4722"}
{"id": "train_035616", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7f1dd23e34cb580d1661ba650a88f29ba50b8351"}
{"id": "train_035617", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "5098cabae6575fd748f6160bedde1548d5429e22"}
{"id": "train_035618", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "eec884bc458f0dd08436b6352c21a5727708da2d"}
{"id": "train_035619", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "5d6e26488ff816bf70a9196586fba0a7d17a2d14"}
{"id": "train_035620", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "279df1dfa497167138ba8e13f7793480ae6ee8e1"}
{"id": "train_035621", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "938264fb1c313dd8224a9741a45cb82a4968491a"}
{"id": "train_035622", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4b7ec8537200358fb69577fb16ea23b108649a67"}
{"id": "train_035623", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "23e8c4ddcfee7779193d4edab6a7e267ab021373"}
{"id": "train_035624", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "7476416111778e2e46e4627b0019a7264ac6c528"}
{"id": "train_035625", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a2f1c0e5f5654f5f41e3a660e9b12f544108a81c"}
{"id": "train_035626", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "69d6117a3f0567f94a125bf5daf8be78bfa3fbc8"}
{"id": "train_035627", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "70385b2087c8091f2cc06b4eee3c209eba0a50c2"}
{"id": "train_035628", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "8bd840bd143f80cbcee53a83b8eb4980529131a3"}
{"id": "train_035629", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "69d6117a3f0567f94a125bf5daf8be78bfa3fbc8"}
{"id": "train_035630", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "d6ffd311a533efe18ca310c1cf3100bcb9794a94"}
{"id": "train_035631", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f62f54e96c37d1365f3d434396ccd21e9c2b0cda"}
{"id": "train_035632", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0903423ab4f34b55e608c28116ec4a9db29ab30d"}
{"id": "train_035633", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "1b858b4a13c1b915b47be7b1e46be4b82f627c45"}
{"id": "train_035634", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "76b26a2938ee9f394c7a89500838ac27a453c766"}
{"id": "train_035635", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d648a03234bd8468ab72e3d5ec932f07a8887e56"}
{"id": "train_035636", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "c8b4b9d89ef5f7643a3ca3d0e2e0bbce33909283"}
{"id": "train_035637", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "9877c5dbeb556c1badd34f5b405f2688b21436b9"}
{"id": "train_035638", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "7c8f79d4ebb6a3a1c3fe06577ae25de3293172f1"}
{"id": "train_035639", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "5e7153fc6ea64008f8c89da9161c7779c92349a7"}
{"id": "train_035640", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "fe21e290dc30f92aadae321fb61cca9e3e948d16"}
{"id": "train_035641", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "5f9ec6f093361f4a1b5ee153be5b7c8c5200836a"}
{"id": "train_035642", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a467f9405f747aa90cfe19a942e3fb46135e4530"}
{"id": "train_035643", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "550c3f41e42369542a16945aa7ca6a69d919866e"}
{"id": "train_035644", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "447174a7cb8c0cece36fc464d0d688adf5fded6e"}
{"id": "train_035645", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c0035cc91d5a4807a81610cb21e54e6ba230d995"}
{"id": "train_035646", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "98e4bc2961ebcfc87b9fbe391db3d7c8496f0404"}
{"id": "train_035647", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "958ebe02d8bda67551b65f0faaafc3593e484088"}
{"id": "train_035648", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "5ead3e9fcb4a1a1aefccaa48955899b520284c7f"}
{"id": "train_035649", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b596ff9b0af62fd2d8c5bb3744f97b672c553ee4"}
{"id": "train_035650", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c88bf5c6f234c6eaf24eb988e13039c14529ea66"}
{"id": "train_035651", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "c7fe978ab0487b355ba94ee092d071f58ad5462c"}
{"id": "train_035652", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b43435b1ce1e7f9c0a79dc13c08485ca896d01ad"}
{"id": "train_035653", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "08199d6d4bc2e88f6bb3ebca70df9014a29e2f4b"}
{"id": "train_035654", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "09cb15711202218c2843e9bf623c904080c3369a"}
{"id": "train_035655", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6a9bfa92458d830be232f40522c46cf2b0cf6698"}
{"id": "train_035656", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "95a9c0993f90565b67341948a14c536c4dacb0c9"}
{"id": "train_035657", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "49c73d1159195c419099e02352fcb20bfcabec12"}
{"id": "train_035658", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "052027b9168f83005ef5a76cfdfcec76c5f894bf"}
{"id": "train_035659", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "11bd2b4255e7a8ef6d2a7ca27e6b99bedf72cee4"}
{"id": "train_035660", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "427b44c988d1764b9f2128e07b0c22e5a244b9cb"}
{"id": "train_035661", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "08bfd42abdae72d0302cb9eb0d7bca16b5edaeb3"}
{"id": "train_035662", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "f50e7693791c94c13ce2d0f5c9f37ecea4a63b0c"}
{"id": "train_035663", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0b39f5be634b09454f3910634d877b1a21bb6b80"}
{"id": "train_035664", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b0a4a6b51bf4fbad49e0e1bb02b84c9f3df7e92a"}
{"id": "train_035665", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "89f024b77d96f18ee229a67469d09e22375405d9"}
{"id": "train_035666", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f8298d41c50401d373cc476adf9f1edfcf28353f"}
{"id": "train_035667", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3ef9fd2f34c748c9219917a6a128356679dc2093"}
{"id": "train_035668", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "32c63f69bf9a5dee88199dc4889c6b73f8733786"}
{"id": "train_035669", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "33a9bc1160df62e000f9577a05fdeaf19be57f42"}
{"id": "train_035670", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d7f50e654e80e6b53b5c14ce682a7a32e2ac9711"}
{"id": "train_035671", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "074e14802ae646c11605ecdd000e5ce5283e945f"}
{"id": "train_035672", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "fa1d1bbadf3c43b9fd094e3020ffd83e4c859558"}
{"id": "train_035673", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "e59680cf1f12e89264047abdcb79bc7de45d050d"}
{"id": "train_035674", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6a44646789349f01ebce65abc15eea70cdde27dd"}
{"id": "train_035675", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "51936bda196d42c3cc62212e3b07fd3a0482a671"}
{"id": "train_035676", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5437e08f0fd7c1213c6310b1d40333886c44caf3"}
{"id": "train_035677", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "a05a3280fb14ed280de5e5cca67d6f95dea4d65c"}
{"id": "train_035678", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "da287370cd4be8a69b94240c39c90493f4d986d7"}
{"id": "train_035679", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "23fbcd5110cebf9c17cea9152e98e18c10b43b55"}
{"id": "train_035680", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "61668d63f33f4aedb942686baa64a78b40e54df1"}
{"id": "train_035681", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "eacc471a5888aa944cf3eea07ac9937cb6ef7598"}
{"id": "train_035682", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "075c2ad7c4254d5d109e95b43eda07d9a9c59f66"}
{"id": "train_035683", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4ab057e5ff30e5b9faf743eded7eea42c88c71f2"}
{"id": "train_035684", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "de276d272c2eb3b0432af1617ec461013da99b1f"}
{"id": "train_035685", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "d48b537968ca4da9924f0ff8df7560bf8a17e0b6"}
{"id": "train_035686", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "fbb3e71a47a342bdaeb737afc21d41e1385e242e"}
{"id": "train_035687", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "275a2e5236c29c54fc9ead8a4f8743b2f4e45543"}
{"id": "train_035688", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "2ae9ef6eff7419ed9dabbc91bb78047b0966effc"}
{"id": "train_035689", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "dadac65941a0a6261592fe38c25ac631644bcc20"}
{"id": "train_035690", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "b7676cbf0451df26ed6daf718d18e1b181a1a95a"}
{"id": "train_035691", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "438aac7594eb4d0f3c778d76e3f839fff8e46844"}
{"id": "train_035692", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "f269ad566856cf4447a2ffadfdc258677d70170d"}
{"id": "train_035693", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "874dedfab530fb46fbd69b0bbc7e2ec7c3272cd8"}
{"id": "train_035694", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "0a29283c994dffc01f2ec886e69a0d1704c7aca0"}
{"id": "train_035695", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "966d03d39e8482a7e6932a476f022378878ea637"}
{"id": "train_035696", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "af0df436744dbee227025602be0c3fc3047a1af1"}
{"id": "train_035697", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3ef9fd2f34c748c9219917a6a128356679dc2093"}
{"id": "train_035698", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "974c1db6eeb2e9386dee94b780ae18837af1d69e"}
{"id": "train_035699", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "1bc9ae1d3cd2e53bafd9091cb32a8e3b6976286a"}
{"id": "train_035700", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "71207c1325f9576beb398d377bc2bfc3afe1c22a"}
{"id": "train_035701", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "a590c8128393f198c5a0c7435d67951d692c8bb8"}
{"id": "train_035702", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "1687d8308f5426f659d5c20d850d1a9debdd3d16"}
{"id": "train_035703", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "e4d788888a005e64670912c4f7a42edb962cf6bc"}
{"id": "train_035704", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "bdb8767fe0bd8a9a205a0ba88398b8911a404d1d"}
{"id": "train_035705", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "937a91312495cde9aa5ed36e8590ba03113095be"}
{"id": "train_035706", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "ce34716cb54cb26a20481625aa36ef42d5436299"}
{"id": "train_035707", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "7ce23915fcfc95fa086bf65e15328cd8a2a2dc39"}
{"id": "train_035708", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "b7541bdd161bd7e07ef596f21da17e657862aa11"}
{"id": "train_035709", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "4b71c2563df50279ae63a8b18db995608a6e7750"}
{"id": "train_035710", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b5296f81dd07d090ff0097428eaac3fb41bb14e0"}
{"id": "train_035711", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "91256f7969bb765b60284e8b08710de18f114c7a"}
{"id": "train_035712", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "01a8d061a65d3a5e63ea994ad16c07cb3a7279c8"}
{"id": "train_035713", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "7b4f4a94580028d2e2120365e449dbe66178a045"}
{"id": "train_035714", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ad4849b876e119ebbfaf8dffcab9b6bac394090c"}
{"id": "train_035715", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "af05ceed642fc682e6be77a7ed04ae53148bbcec"}
{"id": "train_035716", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "6692b5a5fdcef32aeed96e0fc4723fbbd3e4c00a"}
{"id": "train_035717", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a4e8aef6754fa07697ee9735c3e3948d9cefd40a"}
{"id": "train_035718", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "45ad3f39291c94740226cb1d1eb8e33253d7abc6"}
{"id": "train_035719", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f9c9002ea90aba6a480c4b6f9a27b9b4c35941a0"}
{"id": "train_035720", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "f007f2995b7ceac2f73016122d09f279ab015d51"}
{"id": "train_035721", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "cf53a6e5195b395b5909cea90e203a0ed153e520"}
{"id": "train_035722", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "cf0e8525fd29816f1f9c4a9c9789f6e58ca1d686"}
{"id": "train_035723", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "1065d36cfdf4c3dc2dc5edd001b072b5e4882c7b"}
{"id": "train_035724", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "226b87d2c56d02dbef25e9349f49a880df152f91"}
{"id": "train_035725", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "5d7f06c6b34e5ec4943b640ff157fc6ce99fc77f"}
{"id": "train_035726", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "14a754e74908f1ac461bf5292ae19030013a908e"}
{"id": "train_035727", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6a3241fc19a18814736c053d2fee58be00db08ef"}
{"id": "train_035728", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f86417f347ceb4a46c68fea268ec8cb52ca88cae"}
{"id": "train_035729", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "03ff854c0bde68edc5ed18b718f154417a3a9bd3"}
{"id": "train_035730", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "988a0df66d6a24ddcf6f881410e1b2ca9636c7c3"}
{"id": "train_035731", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "6003489d3453b4fe520a3a49a34388791013f071"}
{"id": "train_035732", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "078d988d77765103385f65cfbeddabff2871da88"}
{"id": "train_035733", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "813b53ffb3f5b8027ffec87ae9089bd57ab4f615"}
{"id": "train_035734", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "c0d06ab8491030059be2e5d4410b93ab7efc0301"}
{"id": "train_035735", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b969b5cb3b9df5b3c27d654748930598e48bf1d7"}
{"id": "train_035736", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "58c6669ff477aba863320ec2a040a3fade6ecf22"}
{"id": "train_035737", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "cce81ba46fe0ffd603ec76385f4d07a244b219b2"}
{"id": "train_035738", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "030127799917a8eeb65509c97c01b9fd283acecd"}
{"id": "train_035739", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "dcfdece8466f01bd6a94d90c25c889a98f5cff1b"}
{"id": "train_035740", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a6d3c7ddebd6be396f9d4c8473f15e00b5b12d9e"}
{"id": "train_035741", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "85344f2e43d87774dafdd66c2c8a5f876627f2b5"}
{"id": "train_035742", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "ba8a8200a7ceb2378245b6c19d0bf2635454b6aa"}
{"id": "train_035743", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b27318f2d3abc51d92868f89b64f6f48745566bc"}
{"id": "train_035744", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "df8991d850f9093c9b00323c892d02c050d33673"}
{"id": "train_035745", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "296ae91328a36484b3ac1363e3ea2926872cd483"}
{"id": "train_035746", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["performance", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "78e15df2b90989489cbee52809e8a2cdfe7fa916"}
{"id": "train_035747", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "3fdce2327148066459b02a004bf8ce82ba9ad5b3"}
{"id": "train_035748", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "caf97ed4fb5e0cf5ac5d3282a79ee4bf30db7970"}
{"id": "train_035749", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "b4e6fd46c7b96fe3848fb6ef768ac5892320ed02"}
{"id": "train_035750", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "15443f7e17a3d51373d025afb51a48eff1e53e4a"}
{"id": "train_035751", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "9edffdab2dfe17bdea443c6b673ad105cb23ce6f"}
{"id": "train_035752", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "7c774d1c177c385c6357a123c78c7f02b9d4bb88"}
{"id": "train_035753", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3eee8aaadbd3e750b1dafd2c3ac8ae93870ebe5f"}
{"id": "train_035754", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "0146d43e27a02c79f6669d881a73462fe84072f3"}
{"id": "train_035755", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ab708636a7b1bf09de7ce2f2fa7ae1b5180b79da"}
{"id": "train_035756", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "19bff1069dd84304089908959a1c2195de1b25e6"}
{"id": "train_035757", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "d53be11e46c9793be4307888ae9aca0cb087007c"}
{"id": "train_035758", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "de24b3c34a11c2de006d9eed3518dd068a6fb5ee"}
{"id": "train_035759", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "bdc5826598fa896af658672efeea68e21de27938"}
{"id": "train_035760", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "02c35d75a58f6b5e1f4a530da7c7325ef4bdd995"}
{"id": "train_035761", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "f29428e62b652ae5cc2ee91762924b9b5b34fbc7"}
{"id": "train_035762", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6c7559d76ada76b0e1cc3922b6156a5388d74d26"}
{"id": "train_035763", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2a2b92cf94a364801c4417f52c3047b6e2ce358a"}
{"id": "train_035764", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "7e97f658799a72a35b3c60de1a47b3de95034df7"}
{"id": "train_035765", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b2f9d9e37a0de78dcdbfbeb887c61cc09ae50a1f"}
{"id": "train_035766", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "d189386bfb30d744d014e58121bdea3b83869ab6"}
{"id": "train_035767", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "f989ffa6f44c844fce6fa28368d2f425855534d5"}
{"id": "train_035768", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "2b4fb4c84ce181cd4becfc07c9ceb5af109d3a74"}
{"id": "train_035769", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b3decc8c291975e1c84df6d12270e6db18300a47"}
{"id": "train_035770", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "02740d1df984a6decbef5f1adda62039dc6a5467"}
{"id": "train_035771", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "63fb30be08631884afd0d736b3efb59ba177c9ef"}
{"id": "train_035772", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "42b50fc4714216c3ed6384444d4a13340621e15c"}
{"id": "train_035773", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "461e3b03d90d8e6b75d428eaf84aa4f22b048bac"}
{"id": "train_035774", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "659db7d4ea2f9a3a315ad0caca0d989a6eda40a7"}
{"id": "train_035775", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f7d667e388e20014756424eb8d1b6adba2dcb5a3"}
{"id": "train_035776", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "cec172dfa3dc12f0136923e5bac01b15030830af"}
{"id": "train_035777", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "8fd0295b9648940ea6b1a76d4fb2d4f27c996a99"}
{"id": "train_035778", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "74b1b5f766a335a9ead6222010aee73a4e45875b"}
{"id": "train_035779", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "6d5904c5e895a949dd22b423769665c6863dc4bf"}
{"id": "train_035780", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "5e568295c03f17cefe5845316120b69868990146"}
{"id": "train_035781", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7dbdef615cb2e994d20af6d5591d3f953da0c56d"}
{"id": "train_035782", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "18e9d61665249cbe98df7a8dd4eb34c5c94617ec"}
{"id": "train_035783", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "1c93f9ee85ce5090b9517aef02ef872b8e4a2a0b"}
{"id": "train_035784", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "200fdc2414a5548087b9365c48c6699f6cdb9ec9"}
{"id": "train_035785", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "3acb2df603e226a15768d0112eaa6908c798a4b2"}
{"id": "train_035786", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "dd7303e10a88296cd8e0dfe151ceb8355596a49b"}
{"id": "train_035787", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "be90dcc84e3f36cfc7704e1b3ef7644545831089"}
{"id": "train_035788", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3a9e996bdae2bc0af778ad207bec77b726261c6c"}
{"id": "train_035789", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "6c94fa32f675484767652ac11335baf5aac3a22b"}
{"id": "train_035790", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "169560cd9b0b0ce69f6193d4cc2b5a3098baed0e"}
{"id": "train_035791", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "5536268005bfd71db011413c93558a134661e8de"}
{"id": "train_035792", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "84bd4388c2f92075eb488c704e0b8a2008d61463"}
{"id": "train_035793", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f9300bb8df88bbe9e24c655c5f08851b6527dc93"}
{"id": "train_035794", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "1500640766415c5c43ff74dbfd245cf4602dfd19"}
{"id": "train_035795", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4813654eee2fdbbc6594e4bb772cdab2ee0e5ef9"}
{"id": "train_035796", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "eec884bc458f0dd08436b6352c21a5727708da2d"}
{"id": "train_035797", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "25d90f6ba63dabf6eb44d55e836ef6281e09f1c4"}
{"id": "train_035798", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "dd3a065c0b6effa8b3e91e29d8e7a171a4c0f4b5"}
{"id": "train_035799", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "36d5f92f1120ecdd937871ded71c3eeb5e4182f4"}
{"id": "train_035800", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "02d1c7ae9e22f13f60210f5975dd0f10a676c052"}
{"id": "train_035801", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "5732f75af6592d5a3cdc49a533cdbaf21ae20ee2"}
{"id": "train_035802", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "046e73f9591a3d975af71d9c856f8dead69b7c89"}
{"id": "train_035803", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "3fdb26dd14197d6a93d7e80a11880377c6475ddf"}
{"id": "train_035804", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "6454c74b02acd946388ce583976c51a3421fdc80"}
{"id": "train_035805", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "2b52059ab5ffb3c0389192de1d8bd021ed28b81b"}
{"id": "train_035806", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "0776023f740e0efbb708f207a5a561af09a79817"}
{"id": "train_035807", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f50df29dca65bd8b79c6ffea49c9c23987a07846"}
{"id": "train_035808", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "d06f41767d14e3539944c48d8067ef18101700a1"}
{"id": "train_035809", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "76cb5469752b21e13e3ae1cab1d55e31ab8c98b7"}
{"id": "train_035810", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "898167b565a150cbd559e48052bce7e5f8b1647b"}
{"id": "train_035811", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "de898c104c022d1eb247a75d3be50946e15b2d83"}
{"id": "train_035812", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "9b55b285de03f3b715f67abc1d1f496d95fd5e71"}
{"id": "train_035813", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "09bbafba614366230e5e2eb036ec6bf175e2f320"}
{"id": "train_035814", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "a32142f6cd6095f6355791483e4e6987e090634e"}
{"id": "train_035815", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "788c7b7a6c4d66247ddd8baaa82b93f264f62f8f"}
{"id": "train_035816", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a2b377e4dc6e3df99472d8878a7f497b0998d590"}
{"id": "train_035817", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "6feacc8e4c0b3af3becbe550f9afb1342c64edf2"}
{"id": "train_035818", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "753361d0fa1266758df57c3dc152b82236ffa081"}
{"id": "train_035819", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "43f19e773e9e5f1ec8b8c4ef47e6d6b1343946ac"}
{"id": "train_035820", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d53ee6e5fff4ed52804da1539e1d83d503ef9d6d"}
{"id": "train_035821", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "7aeaf9e9fafb5b1d883926d80b060e410447a6f7"}
{"id": "train_035822", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "dcf22b3b82ceb1d86a8ebe8f32b45d5f0e31bf9b"}
{"id": "train_035823", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a6ae2c946cc9d41b7cbc8d7681d302556546a366"}
{"id": "train_035824", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "6751fcda7a1a262a66525e74cb59b82f35a64263"}
{"id": "train_035825", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "918a662f54cfc8698fc15b6baa0de1d989c74a16"}
{"id": "train_035826", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "bead31dad8c126dbe8fdc4329b101e281065b3f8"}
{"id": "train_035827", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "efcada577410cf93b7ad13b422a8146091afda56"}
{"id": "train_035828", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "46b9c6b0cd8fc6d86952837db9b106cbb5c126c9"}
{"id": "train_035829", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "558c10bccdcf2f4bbd644b4ce78154d2334dfd50"}
{"id": "train_035830", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "6043a3c28091f5956c67a4b79bce596a7eff8a72"}
{"id": "train_035831", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "81fc6389df841c467696b92a313e6d358cac2146"}
{"id": "train_035832", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "67bfeb5bf076ca579530245505e80a1d3469ef92"}
{"id": "train_035833", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "128b26e7038fecdead43aca7cc30d66ba9dbd9fa"}
{"id": "train_035834", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "354fafb4890c3cb2a5a413fe5c9c9ecbe8b32be9"}
{"id": "train_035835", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "0a5c738fb38d50433dcaa2bdf36a7633445a88f9"}
{"id": "train_035836", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "30fd79be8672564062c881212c0d77183c8fedce"}
{"id": "train_035837", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "290d97891e1e978141e29ba7731cd4985665b448"}
{"id": "train_035838", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "408faabe26f99a40219a91fee5b31df1b7ef27ac"}
{"id": "train_035839", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "50160b93fad4399984fe49ca71f2f9509c8b01f6"}
{"id": "train_035840", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4ee394358beb71c07d45a0679917e5fd23cca195"}
{"id": "train_035841", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "350b4e4b0d643944ca504a426b990e90631768c9"}
{"id": "train_035842", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "899e17fc08fb8c98778b0e7617b6f2ddbb229f8f"}
{"id": "train_035843", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "bbf7a46af949deefe3a9ae954e7e64562763cdbd"}
{"id": "train_035844", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "8c784d393c1af6619ec32b2c02561c82a1a245c5"}
{"id": "train_035845", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "eaf09b41539299b2db2d8353f22670b6fd5af487"}
{"id": "train_035846", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0a2ffd79e8d0a4a1dfa1a95d6731de6cf60eb2b7"}
{"id": "train_035847", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "3970192374e22a757cf6cac98e3f2ad181950905"}
{"id": "train_035848", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "25ebb975774ca886203144900532c9d176de92a9"}
{"id": "train_035849", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3cc9cd2e13661fec723b5251c26f36c67d60e8d9"}
{"id": "train_035850", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "507fa7c523bca59eee7aeb7100730ede127c86ef"}
{"id": "train_035851", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "bc8ba07e8f2f4579a9184a08ccd0586a2d0c33a9"}
{"id": "train_035852", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["performance", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2dac9a66e6995158a860b0e574fbd38f8b11d2c7"}
{"id": "train_035853", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "69e41ef4030cb0463b9daaf039990dc3e5c38254"}
{"id": "train_035854", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6adb4b71cb02d8bc272da9686d2645a27e1a006e"}
{"id": "train_035855", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "bc233da0b8b6f4845e22a38275bbe41e5e217ee9"}
{"id": "train_035856", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "384cb83d6350a86fde6a5eff93d85dff8387f76d"}
{"id": "train_035857", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e26f15ed58edf3eab2403402c0dc5b115d48823a"}
{"id": "train_035858", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "478f372f4e6edefa72ede1565c81d6664de17a75"}
{"id": "train_035859", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "1ad4b8d264c4427a4322afce6f5e2367560d177d"}
{"id": "train_035860", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "d9ecb555c83c1c0c6bb8e3cf582c93d7790ea676"}
{"id": "train_035861", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d9d1494b6ae424641a7faa86950bff8c7d9907d9"}
{"id": "train_035862", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "588f4546bea4cadd2de462f3763e25a7904c7294"}
{"id": "train_035863", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "59c7a2e54522d32c4d2395ae908c8d98a0b0fcaa"}
{"id": "train_035864", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "cbd612e6c6caf25940fb3baa90b9f2af24955d35"}
{"id": "train_035865", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b9228577d212b9f16e299f490e6e889b60dafbb8"}
{"id": "train_035866", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "5aad104eabd7aff882d76ea4dfe1a7e9ef0ad04e"}
{"id": "train_035867", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "162a002e3a3372573ff822b4c87bd7c90d86a54f"}
{"id": "train_035868", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "61f5b952830451d53ea49cacaaea99c7ecb9721d"}
{"id": "train_035869", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a57e7c2a6c62a60a20bb604d6ef6f1270c2e2316"}
{"id": "train_035870", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a7019b72bf03640e650472c1db1b928b44389be7"}
{"id": "train_035871", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "1f158bf72f926baa5f3ef62fc33e8fdc22a4d8e2"}
{"id": "train_035872", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "34b421a63096c27d7c6197fb94c5cc84ed06eca8"}
{"id": "train_035873", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "562245a527efdcfb112d9a7a4ffd63c03d679e0e"}
{"id": "train_035874", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "7bc99936c29dac720211bb4c495873a7e5c2ab2a"}
{"id": "train_035875", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "960dbc67c167704834089341c6b69802f7c806ea"}
{"id": "train_035876", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "6788595696d12eb3386295def7611e173310ba88"}
{"id": "train_035877", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "4dd0109fe6d2940e1f03d29af2ebb0ac2b04e999"}
{"id": "train_035878", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "e514092845226633efc4fe820f92c7e2f8eb2c53"}
{"id": "train_035879", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d8919b61a630d249a5e805d7d5b59cb05d9f3a7c"}
{"id": "train_035880", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "dbd3f8162f4ac5310355053e437c3eb5b5a2b82e"}
{"id": "train_035881", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "03977ec0e2e52efbdebbdc7de030187bc9786bad"}
{"id": "train_035882", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "db56baf1597142c8ef0d46c75528c91c14e61ab3"}
{"id": "train_035883", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3eee0f5ce6492e935f9fe65a97ec54047d27445d"}
{"id": "train_035884", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "7c631464588e0c79490f3416268d74bdd5ad5796"}
{"id": "train_035885", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "10d1f54406c8953d4bce83567625321b63bbc46d"}
{"id": "train_035886", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "39c2c5d3595e05e8c1387e70791ee796c1258df0"}
{"id": "train_035887", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "114e2828f742477bce74698408d532a482dcac9b"}
{"id": "train_035888", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3fc523633a9f0fa79e7cc17c80a27fbbafb7b01a"}
{"id": "train_035889", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "ef79b89d9ee03b218457d5504fc44dcc974514a3"}
{"id": "train_035890", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "5ff4b38a5cd2525b40d91e9bd81cbf9782d0dc86"}
{"id": "train_035891", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "775f8fd200146cc609ef804426fa075e8c0ef4f6"}
{"id": "train_035892", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "26ccc06e63a077472482b0d39e2ee4289eedb19b"}
{"id": "train_035893", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "534b4df7a648e1162342e164439c5288f53224d4"}
{"id": "train_035894", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "24ea9f77baba9d1cb8c816d96a09f5c0e2e6144d"}
{"id": "train_035895", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "80487feef9b272b7f8aac9330a068c2c2435addb"}
{"id": "train_035896", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "16871cef1acf38ace4f9e0a1e304ce0fff5528cd"}
{"id": "train_035897", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "6e701e10f854bcb6041d25dd13c3a09c697220fb"}
{"id": "train_035898", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "dd3c2a0fe0b35920293242533c970a9165decd2b"}
{"id": "train_035899", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "bccced7dba4ff6b2ddc5e6ff3b6ef2100325d44b"}
{"id": "train_035900", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "1c597cdb234ac747232cf03391bbf9e18f7160e7"}
{"id": "train_035901", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "aaf4f32c177c58662d58bbd6ca5e70dc2b17d6d7"}
{"id": "train_035902", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "01bdee23895e939d6fcae80d6c6770f652b8472d"}
{"id": "train_035903", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "08550245ff41b98455ca3ab6ef7b1113f39aece2"}
{"id": "train_035904", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a29a68d2bb5c4f20f60b93305f9d262f3c6a03d4"}
{"id": "train_035905", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f29dc82233b6e37ce56c825bd19ccd20e4e36e15"}
{"id": "train_035906", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "e3d723e8707a381edd462b7200cf2930aa8c86ff"}
{"id": "train_035907", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "f7f17b0dee8e1992fbbad53d421ceb6bb05ea70a"}
{"id": "train_035908", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["evaluation", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0aed563e7b563e71d539d8e44da227dedcc9c18a"}
{"id": "train_035909", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "95f335dc0c41bd69be4da9cdba767bab76c9c6dd"}
{"id": "train_035910", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "ad5070e71080fec961d632d2be971289c10c9f3f"}
{"id": "train_035911", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["performance", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "ff2a98731365e7893b376c0a8ffdadf11d330dd7"}
{"id": "train_035912", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "9942e609ff3913e891367e69974cb2301381c074"}
{"id": "train_035913", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ef1d0dcf982c7f4cb3bbd3fa2e0ad135214538e9"}
{"id": "train_035914", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "887fcde6e63be1e969c9e0467fe29436b1f685a1"}
{"id": "train_035915", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "42bd266260b033157c1aa179b18e5774f7b8639b"}
{"id": "train_035916", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "2416bdce3831058255aea92defda2cee2077d125"}
{"id": "train_035917", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "6cea7d758d2b223dbec508b61c056cb40023f5b6"}
{"id": "train_035918", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "3890b91520a52d4cb2584eab9859467588058b34"}
{"id": "train_035919", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "031ace4e56aa79d17d7e30a4a2371cef1c58f2a8"}
{"id": "train_035920", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "8f3a2d6d882cde7e2f1ba6336bd43e28db8cb967"}
{"id": "train_035921", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "b7d0d0f519ce23a34f452ef1a3e702330c5f5ac4"}
{"id": "train_035922", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "d8577b97b3f05ad0667926738645502dd40d049f"}
{"id": "train_035923", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ccd767d2e6583d120a063799db4920cbeece8f89"}
{"id": "train_035924", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a40931b82e283e90bee6a6553441ca75588b6bd4"}
{"id": "train_035925", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "2b3fc26b3428845c6aead6b3cfe059f079caadd6"}
{"id": "train_035926", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "aee5fd62cbd2f7d8c31f457b81a7d848a1a27a3f"}
{"id": "train_035927", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "943aaba7a371c1fdb692fb087bde346f888145ac"}
{"id": "train_035928", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "de5c1ffc5816caa4807516d59f2a0725333b4f14"}
{"id": "train_035929", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ac59ff758f563480600e2d00879c89367ed5e5d6"}
{"id": "train_035930", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "321037c800d208ed6595677da705466abae5e529"}
{"id": "train_035931", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "49c0b5d5b2dc00ac6f66a9066c66043ec31c6118"}
{"id": "train_035932", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "afa6e9a5970d52d1b1b636ef9f6af777dee133f4"}
{"id": "train_035933", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "c89774f32a51dd2b23b397dd6cdc021557a0e5ff"}
{"id": "train_035934", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["research", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "de12ea4bba40c465b98612d1b02086dbe8613a22"}
{"id": "train_035935", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "75acf7440a1e960e4f32a960dd05af6060342742"}
{"id": "train_035936", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "003247c03aac727caa81619533398cc0ca24415c"}
{"id": "train_035937", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0cc615f989e17d797951782cab4d936e926fe873"}
{"id": "train_035938", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "036f832acac36000175f7d139b6d8f5b2cb4d21e"}
{"id": "train_035939", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "3dc0cdbe9491b6d2d0bc727b0d98ffe57fae6cfb"}
{"id": "train_035940", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "58c816fa1c972ab11fd1a6e924a234b84fdf9483"}
{"id": "train_035941", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "388929d7d5a69d45ff0ef9609b6e51d36352b27d"}
{"id": "train_035942", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "607f4201527235a9a6223d65b1cfa29bdedc58a1"}
{"id": "train_035943", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d06a7f42151d7db2a4352cc539582093bc59d1f1"}
{"id": "train_035944", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "2ba15d983438e15ae2de5306e01e1d820641285b"}
{"id": "train_035945", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5a1da8a0df10fc9674e99e49e242a53792f71d46"}
{"id": "train_035946", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "106219da6843a4610641fe7dfce7d62e341e9b13"}
{"id": "train_035947", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "cc322d4ad5ee3b49d78e8fccf32cf5ce0d878615"}
{"id": "train_035948", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "14f4f6b84fdeb73222ad7bae75a464d78135040f"}
{"id": "train_035949", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e785e16998e7f344b33400c0d953940b68d68086"}
{"id": "train_035950", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f973d2742a5aa5776bbfca657da41655ca00c5fa"}
{"id": "train_035951", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "8320f99346cf0ad13c14ea538193f1ee19cd7dab"}
{"id": "train_035952", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d60bc1ea948006d53dbfba99ed1d7ac5002c02a7"}
{"id": "train_035953", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "b321806018e14cfb7423495c2c7fded660b1b917"}
{"id": "train_035954", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "29c74f455b9bc55023b69993801a958121796087"}
{"id": "train_035955", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "1daf4c33a73085516710623d387c394e5afd82a2"}
{"id": "train_035956", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a054ecf33b1329420e6809fce079859eda4fb039"}
{"id": "train_035957", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "3e409f039b3fb158be4413da98292b3536ef5b10"}
{"id": "train_035958", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "429626b4a4ffd06edef6779e7ab2cc272981015d"}
{"id": "train_035959", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "7f2f6da07fcc6103b7535aebb9d2c8a4541681fd"}
{"id": "train_035960", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "796f55becbd3eac9dfce1232a6b3ed2221cee959"}
{"id": "train_035961", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "c9a9dd4f9ad7d54cfb917dfcb21bc9526e43b335"}
{"id": "train_035962", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "05d0cd62b8f3a3e6f7565f2f8d74d82406d8eea4"}
{"id": "train_035963", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "2ced7a6f19dc3504f04615d2a4cfafff4853e60f"}
{"id": "train_035964", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "b564fe25751975ad6db2cc331618361ea2d3c809"}
{"id": "train_035965", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c66b1908b068c85505619d46608c0f3f2887136b"}
{"id": "train_035966", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "58637139f076ccf715371fbe9982b7e9bfbb6892"}
{"id": "train_035967", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3564d59148c9b703a6d96610ae6c464b79b69c67"}
{"id": "train_035968", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "80487feef9b272b7f8aac9330a068c2c2435addb"}
{"id": "train_035969", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "3343b62d848c0f9c3ba90789975e6cede84df78c"}
{"id": "train_035970", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "915afac4832676a45d9ba2394b15a18254252183"}
{"id": "train_035971", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0cc615f989e17d797951782cab4d936e926fe873"}
{"id": "train_035972", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "822adbe96024ac418555e24f1d589a30bc2aca01"}
{"id": "train_035973", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "5f43a2d864b3109a5bd7fee88596227d201478e9"}
{"id": "train_035974", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "fa0e03789772b9bdfc1951b02981252984b6bd3a"}
{"id": "train_035975", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "214fe11f103cc850fad353c3ab4cf903e85fb857"}
{"id": "train_035976", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "0f69d8be694aa09911c9e20a4609d6a39955370a"}
{"id": "train_035977", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2772d223a63a7da4a1a4cbde6efa7b8aeb5374c3"}
{"id": "train_035978", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "9c2d1fb224630cea2007058188358defde82ec02"}
{"id": "train_035979", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "97fb0165c60f96a5519a6ce3150e88bccd24d5d1"}
{"id": "train_035980", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3a073500ff6b8a1efa524f51c57c51b743ac7544"}
{"id": "train_035981", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f205f4db6a06c571548a2fe5225dd09dac616910"}
{"id": "train_035982", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "22980b0fdfd5efe75a15f2a1f88c0cf7ed0eaf94"}
{"id": "train_035983", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "673a5c3a70edde8c972b7d233dfd399d3349f0e6"}
{"id": "train_035984", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d4224f7210d9467d5fcfb806a7c307ff929bcd7c"}
{"id": "train_035985", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "23ecaae1f12ce138c7436b99c8d17c5a16617099"}
{"id": "train_035986", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "53c20af2a58df865278778cc8599dfb4ec3dbdd7"}
{"id": "train_035987", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "18101423af537a879cb5a3d1ac4fd5eee2d6d436"}
{"id": "train_035988", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "ce19da4006541de060751c9c6d013ee2130ef51a"}
{"id": "train_035989", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "cb2136ccdce9db0aec891e05aca32004627c3c94"}
{"id": "train_035990", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "ad1e0ce67f3e222d111bdc53d2439738066e3209"}
{"id": "train_035991", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "49027c5e14bd07896da42c43d80b60f37be83631"}
{"id": "train_035992", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "239a72b4015759ba680857f0e9a41cf038172e1f"}
{"id": "train_035993", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "9b633dd64d83f7a2d03b92536cf9c9e9185fcccc"}
{"id": "train_035994", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "1f9a0021ebf055d1ada8ff8b48dd676dc9c4896d"}
{"id": "train_035995", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d395a6dc6b5e0dcf2436776bacc2d367812ca24a"}
{"id": "train_035996", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "4713cb0d064c7720c014b804400e63f312b65dc1"}
{"id": "train_035997", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "74a9f4d65997865b5e985fd2df4cb6df9fe995f9"}
{"id": "train_035998", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "ed3c0e0e5219424ae82d92756d317fe035e3fcc3"}
{"id": "train_035999", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "2f54caaeccbce50cc3b4693e3f28912135195f1f"}
{"id": "train_036000", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "aca90055463714a920406d9c8ee0f7c021a7f0d2"}
{"id": "train_036001", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f9fdd0a0d79d1a0b1b35b79ea93d18488556faec"}
{"id": "train_036002", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a25f92c009eab7dcb1369445ea035a50b2836a35"}
{"id": "train_036003", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "de898c104c022d1eb247a75d3be50946e15b2d83"}
{"id": "train_036004", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "5780bdc642b7d9862487a5af05cb1bfb5212aff1"}
{"id": "train_036005", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "f6a2c092c7c86c79daa8aec79089523dbd3bcc97"}
{"id": "train_036006", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "69b4a35183cc10225b85efac66e10af231373626"}
{"id": "train_036007", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3149f56b2ab271a091a2d66338f317a9b50362c2"}
{"id": "train_036008", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "752cee9c668c2d6d27c14fa611984f41ee889e43"}
{"id": "train_036009", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "417cef637f4e9c400daa8e6ee7df04bd32f61fbd"}
{"id": "train_036010", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "8610b88a510e4d433af73a2f38d49160efb6c4b3"}
{"id": "train_036011", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f17c5ba34e019682daf0a4bdfd5185e66e5f447d"}
{"id": "train_036012", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "84b862e3d9e9d0b53641e3f4d3138ab9f366f245"}
{"id": "train_036013", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "c93b2b3db404018f8fc5fe2fd33173d7ab608933"}
{"id": "train_036014", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "8b0ebfdfda7b469ffd42cc43b6600c188cf43df0"}
{"id": "train_036015", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "500e4d9ce3d8edc5c737859ab0892c5371c95ddf"}
{"id": "train_036016", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "939c166a1e4eff1bda26f894bd68a5fbba23dc5d"}
{"id": "train_036017", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "0ac85336095dbd141f9afddf3f16054784799e35"}
{"id": "train_036018", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "8cb0272628817cd8d82669e5b97dd68afb2fc1a2"}
{"id": "train_036019", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "13ab93960db8bb35a4d832be2c634834087fc624"}
{"id": "train_036020", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "3d408bb18f315ce30725f226f3103e3e7addd271"}
{"id": "train_036021", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "b189f6c91bdf64fe9eb706782871b54758275e6f"}
{"id": "train_036022", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2d5f1bb20d7cbd4973e7b80af4e464cf4cbfc8d3"}
{"id": "train_036023", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "2346b99f42bf2cd46400d4f273c7ae7bdce5d5b7"}
{"id": "train_036024", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "75066fa7f3e25e093f5a2f9a50699701b2d995b9"}
{"id": "train_036025", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "716f394920728c66079f43a6a420884b1045052d"}
{"id": "train_036026", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["governance", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "4b2868d4853879aea1373bf58ed1360df3cc384c"}
{"id": "train_036027", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "80bf3a9a0d4e734c18ac31f57d1a52f552637077"}
{"id": "train_036028", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f42fe00cbe59904d5f2f19778b2c7d23f3fbf8bd"}
{"id": "train_036029", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "413270efac9c51829c7fe1217d5fe61bf1083644"}
{"id": "train_036030", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a94f49f2afec2fe25da6a259d33bf8678344fee9"}
{"id": "train_036031", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "b1fe56c2acb168f61822506c9362dc0ccac53601"}
{"id": "train_036032", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3ce2398c9c92b7d4aa1c431fc9a564df2d328819"}
{"id": "train_036033", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "bd25d1bb807db30e2fde4c8c2e33a98906310155"}
{"id": "train_036034", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "9c29a0bce4b19e6838267eb26250c3b5f1314ede"}
{"id": "train_036035", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "17fdfdeaec9d3f77903c88be0237704454a8afd9"}
{"id": "train_036036", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "41683dd1c0bab35a595981a3f28de4289f8b5724"}
{"id": "train_036037", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4ff809520b8b0575e82966420ac0c0fe5e6018ff"}
{"id": "train_036038", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "58086daee3833a2b85c1074c9234a452c6d021e4"}
{"id": "train_036039", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d2fd4554d5f450853b686ae15b7a350973d9b59b"}
{"id": "train_036040", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a92a1ce538a3b1090939668a545fb2f8ce8e85e1"}
{"id": "train_036041", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "547fcff9fc4c4cacb79498987f9d3a77bc570749"}
{"id": "train_036042", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0984e349f86ea65042f6ea51aa9e95c434ddbfa6"}
{"id": "train_036043", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a11295f4563cf8ca9fc933785c3614521fa2ac74"}
{"id": "train_036044", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4fc93d9aef1020e2b643e8a46eab96f09c129200"}
{"id": "train_036045", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["governance", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "beb03ae4be43fefa3abbee516e89a8725b73f48c"}
{"id": "train_036046", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6479f1d1ce203a42d5695cd41bac2170c518cafb"}
{"id": "train_036047", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "09651b664012c0c03c787e3f40bb5dbea7baa28b"}
{"id": "train_036048", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d3572c348e13ca566892595065a7bbe25e123efc"}
{"id": "train_036049", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "12aadfa00a0c5cd91d1c0dd498d0175ae97e559b"}
{"id": "train_036050", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "853969ca7fa1a28e2b1b53cdf8928db57d174161"}
{"id": "train_036051", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "08199d6d4bc2e88f6bb3ebca70df9014a29e2f4b"}
{"id": "train_036052", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "f0b31776383f1f8693e05ae8e2944cb7cf8b7be6"}
{"id": "train_036053", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "be5f754434c5882c7e44040caef7f36d4c221057"}
{"id": "train_036054", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ed4a63031a93f94a3707afac7bf4d3a4503db683"}
{"id": "train_036055", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "5eed11f347c5fc8da3922dbf668164bc38fdc2a0"}
{"id": "train_036056", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "dd3c2a0fe0b35920293242533c970a9165decd2b"}
{"id": "train_036057", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0735a754224c15b9939cd70d5175159188f1f919"}
{"id": "train_036058", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6e3c0947c30a58a0a9a21b2c6f9180024e9a125b"}
{"id": "train_036059", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "ed50fd2bcf67a85113e0bfcbfd8c0b5e09121844"}
{"id": "train_036060", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "8f8205231d0a3b0552126081b9b6b957a74fac64"}
{"id": "train_036061", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "2a372a13bf0bc4b08fa6a9bfb510c1cbb10b6a2f"}
{"id": "train_036062", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "8664758a35521975ad7de481bc74024dd5aa166c"}
{"id": "train_036063", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "85c2af5a32a94e39ca064332e87d44db6c8d11f7"}
{"id": "train_036064", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "aec3b33d50f16f0a7e806c10bf30ae8622ecac60"}
{"id": "train_036065", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "4a6eb848575bc59213b9a48ce86f9c96dd995a7b"}
{"id": "train_036066", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "44e28802d170c4031d20a1fede46f9c8bbea9b55"}
{"id": "train_036067", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b77eb28a3be1ac1de02b584883cddfff45dbced1"}
{"id": "train_036068", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "2fbf5a34a9b31916688547f2a8563a2da50b4535"}
{"id": "train_036069", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "7acc86b16deb68beb32ff0cb6f03e994dbf7107b"}
{"id": "train_036070", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "ebf9703928c7c6ed76da7f7beb3a982246825c3e"}
{"id": "train_036071", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "8df662f2ed01e1e609976348369e57ff037daca0"}
{"id": "train_036072", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "4e485f92a9dc32f84decd65134ba60c448f1b805"}
{"id": "train_036073", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "9ec3b7835d76d1e361a5f00b5940eb2d8d692bbd"}
{"id": "train_036074", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f12811ce14caab05c315e36fa8b4a0c91dcf24ca"}
{"id": "train_036075", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d24c60737921f63b931ec208231f528c3d75e595"}
{"id": "train_036076", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "7f6acc6c0acb1ba4ee8798d5d39bd9c9967117a4"}
{"id": "train_036077", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "614662976a4d4d8333b340b4b38a5f94ffec3651"}
{"id": "train_036078", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d17e04da91d44b998ee96414f657aae01b596549"}
{"id": "train_036079", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "942a43b8cfab0645b0220c6a5bb228f22555a5b7"}
{"id": "train_036080", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "0ad38ea133e64d531f78ba1e0127e79de43adb58"}
{"id": "train_036081", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "88f514e4cdddbb9177232ab4209b22206a476266"}
{"id": "train_036082", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "002fb5656dc3a2d3425099ad3ff79247ebfd9bbb"}
{"id": "train_036083", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5fbcfd5d4ef79cfbbd52c05e9fe716920acf2392"}
{"id": "train_036084", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "1cc71091134498597c0104778ee74c5256f72e86"}
{"id": "train_036085", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "9954005050a413d512faa17e3837a5f49beadabf"}
{"id": "train_036086", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["agent", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "c4e8b8482179e775381d9108c21be514e63f4e1a"}
{"id": "train_036087", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "7b4f4a94580028d2e2120365e449dbe66178a045"}
{"id": "train_036088", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b1491183be337c3552b630d53c4fb7b4e197b88b"}
{"id": "train_036089", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "be19a0ba8659bb34b6c7d0f2a89241eced244f66"}
{"id": "train_036090", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "93a2a267127cdbe13340d3431fb0cc18048d9865"}
{"id": "train_036091", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e1a2e3b20fba4d6d2490067283360efbacf1736f"}
{"id": "train_036092", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "8e31ce798fbc05b0ccf6a70fc15295074820ff4e"}
{"id": "train_036093", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5f816bd91dd57cd2508ae906a1a2da2901c1b88e"}
{"id": "train_036094", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "ccd65bec1a55fe45a2e15a3e6cf2683bdffb5b0f"}
{"id": "train_036095", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "07a2af5e8b18a939d7ae2b54485017afbc65257d"}
{"id": "train_036096", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "5c710b795aaae0ab8eded346bf2ce088a57fd404"}
{"id": "train_036097", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "14da01080daf8025c57a661a26da2de12d9167e5"}
{"id": "train_036098", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "051df7b5a82096e64b216158f0220a48df2afba4"}
{"id": "train_036099", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "4a143e148e6efff664b730f731f536e619dad61c"}
{"id": "train_036100", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "373c027f46b692d7aaf40efe0f776835efb360ae"}
{"id": "train_036101", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "fa42b21d51e434ffdcb0937fe5d8cea57e07c5dc"}
{"id": "train_036102", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ac943138118c347d6006c54166b478028c1d28f9"}
{"id": "train_036103", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "8e8c15786b04d4b864e506efa99b8ea3379c818f"}
{"id": "train_036104", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "54aa271b5814297f72c149199630f218cbe479f8"}
{"id": "train_036105", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "9fcc23cc3a357a3ef002c91f40b263e15b973c45"}
{"id": "train_036106", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "63cfbcc751193d1d7e04dc6b7bc351de9cfe62f8"}
{"id": "train_036107", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ab26f03de059019025c4437e418bb93d5aa44c54"}
{"id": "train_036108", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c93b2b3db404018f8fc5fe2fd33173d7ab608933"}
{"id": "train_036109", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "265ad80743535cdcd3d5dd91a44b5f7e2aeb3af5"}
{"id": "train_036110", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "725825cf1f9fd5a6d3e40b72b911189ab78c58a9"}
{"id": "train_036111", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "38c547184f396e1219360e5e04e16122a1578f54"}
{"id": "train_036112", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "668839246e1805a15ad8045c967f05432aac01aa"}
{"id": "train_036113", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "2ac64879e7be3bc380b6f2dad84b63146267aad7"}
{"id": "train_036114", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "6f88730f0b476057a84461719224fe6f5337b59e"}
{"id": "train_036115", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "9668626a40b31e06e4e0c8bdd54870784314121e"}
{"id": "train_036116", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7a2f3e7d5b34b933fb7b0b339c9f40c496bd347e"}
{"id": "train_036117", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "478f372f4e6edefa72ede1565c81d6664de17a75"}
{"id": "train_036118", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "35cd2fa791fb3a2b46a3bd9bdac6c9fb4c9825b7"}
{"id": "train_036119", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e394d9de662e36203fa06d382d12bae0238fded5"}
{"id": "train_036120", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "81589169437cf630bd75f8fc2d17f00383d057f6"}
{"id": "train_036121", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0496ed9a107d958eeba9ff63b6b985108db1a3d7"}
{"id": "train_036122", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "be9e134c9ac31073aced3a7b7be99f3870c40401"}
{"id": "train_036123", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "0ae48054c05b7f92049dfba1f9fe8efa199ac02e"}
{"id": "train_036124", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "0c01ffe135fc7fb7e93e26461cb4ab52552fc3c5"}
{"id": "train_036125", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "a5de435469925a2a2a5d0553c8eb84b6cf7168db"}
{"id": "train_036126", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["security", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6a17e2fca188d6f58234e10d9fdc83828225d4bb"}
{"id": "train_036127", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["research", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "93908807cbae367ca6cdb47ba29e10ef5ffb36b6"}
{"id": "train_036128", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "942a43b8cfab0645b0220c6a5bb228f22555a5b7"}
{"id": "train_036129", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "3f2daaa9da64bb474faf682efd64efbe3447fa88"}
{"id": "train_036130", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "75a32e7f3b8db0e43b81e22a8a2caa9cd4b99dd3"}
{"id": "train_036131", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d99c092be5d8f78ff99fae2d748a5f98a6a51b6c"}
{"id": "train_036132", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "9bc653128daa38a548dd5ff34e1bd05f242615ea"}
{"id": "train_036133", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "97867fa70f87f499b9c8e825f9c7f4dce5157847"}
{"id": "train_036134", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "d2854a1760b5b39ef174104153e9574d7f68ce90"}
{"id": "train_036135", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "62ce6fa891fde29b766ff4b3eada804d2a629df7"}
{"id": "train_036136", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ca9cd464027af89775add5f9c6e80ee04c9d7867"}
{"id": "train_036137", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "576bf73e55bd2a3dafaa1c3929c2c61b1bbd8b28"}
{"id": "train_036138", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "de79fbe63e06ad2707f4315a27e8b447da3862e3"}
{"id": "train_036139", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b1443d496473e621b0be8bf49d7e8403a9a25041"}
{"id": "train_036140", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f0190ac75cd51393968cdac6fad154e9f5b98a84"}
{"id": "train_036141", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "f64f7fa195cf70de8a8624f14dad13992cd5af78"}
{"id": "train_036142", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "ef73773ce8d321ef20d1f45370f456eef317240a"}
{"id": "train_036143", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0c7cda9176f55f86f7517996abe86ee06cf2a24d"}
{"id": "train_036144", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "eb73311c1af2a0b32120706eef734517f7a4fdbe"}
{"id": "train_036145", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "5233288905d1ee477c217b61273c3f85f00ba1d6"}
{"id": "train_036146", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "817b60b7024a6dc5479c21c2a7e25688b414e60e"}
{"id": "train_036147", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "99263976f32c08074d8b04b3b11bb0db6c9591b7"}
{"id": "train_036148", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "aaec72989924dd3cee917f263d021909273cce20"}
{"id": "train_036149", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0f7fd1235b81d9e434eacde8f34bd46394c78309"}
{"id": "train_036150", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "7ea14f6ee0567321f547bf0813c3ccad7349c8d2"}
{"id": "train_036151", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "6c80ab493f518eab77bc63c407305587d614484c"}
{"id": "train_036152", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "577cabf3ea5d5f1d2c600387583f620129bd280b"}
{"id": "train_036153", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "626000982917b37db06770a6b38eec18f8705e56"}
{"id": "train_036154", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["agent", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "cc24267b8ea39ee7b54891b764cd13c9fcf57480"}
{"id": "train_036155", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "21dcbed354e352d0e5d96ab4e8039c3c866f8c64"}
{"id": "train_036156", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "601582afbb75fe3d919d91d00a9abaf42d24f4ce"}
{"id": "train_036157", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d2e2bb519202e1680081ff1b40597fd61568c3a7"}
{"id": "train_036158", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f5ba595ba17a588d39017f76f7619e01ca760f75"}
{"id": "train_036159", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "6c804547d12284c0df859cdc3508fa022a154de8"}
{"id": "train_036160", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a46ef1b49bc587e228c1121b2fea73bee2f5c156"}
{"id": "train_036161", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "3e382cade3e60b4a74deacf670065c6697854e15"}
{"id": "train_036162", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a44db2c6d3dbff7ac7d16a6e6751d2d5ba4bd322"}
{"id": "train_036163", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "047638ab34e739fa9c22d22386d222daa5a0db1b"}
{"id": "train_036164", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "90a854c6f0714224c123f863ff999d6fdd93b29c"}
{"id": "train_036165", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "17292f089361dee6cd1b7a1fc87c869eabe38481"}
{"id": "train_036166", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "38bfb98829c878b1518a4798d84bb3519354f426"}
{"id": "train_036167", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "ecf57d101fcb7225b1239090bf505098e26d41f6"}
{"id": "train_036168", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "fa187b39eb20deb39785fe15808cc50a4d38efec"}
{"id": "train_036169", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "73e6657c630cf2ef65e46d6950ea096ae211bb79"}
{"id": "train_036170", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "ba4a16d77c1c509e0c1e5b3451cdb691bb4c449d"}
{"id": "train_036171", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "8e8b6f0846b1ecabb12da717dd710b79516254bd"}
{"id": "train_036172", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "4bd2bf51fc40a17d2005a61600641c850a006bd3"}
{"id": "train_036173", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "4370844c43bdf758b7a4966c632081e523f14005"}
{"id": "train_036174", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "34015c77acee6ca81795c2a0a81fb0915416b0db"}
{"id": "train_036175", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "68780ceb0b325e8ea7a9948531ffd8f9c0f66240"}
{"id": "train_036176", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "90e564ce9b818b2227e6f32515eeb54f9a0e782a"}
{"id": "train_036177", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5bb41acdb66fdd6113e4a41d6de05d883d740816"}
{"id": "train_036178", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["research", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "618ad0ee1875d8df538bed5c8e474f8ad874a053"}
{"id": "train_036179", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "a2fe3cfea76edeaded90edb9768366d9e3783a5c"}
{"id": "train_036180", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "91e472a956897c06a3debb098a0b85053273bea4"}
{"id": "train_036181", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "eba3056dd703d14cdcaa30677f0cad11b92963bc"}
{"id": "train_036182", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "5302110f1e0f4ad9a0d143f16c972d56ec78aadc"}
{"id": "train_036183", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "7bc7cd908215dac7b522a7c00a8b4312086c0558"}
{"id": "train_036184", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "d3e63929e17d5eb59090f5775b92b65a0b97949a"}
{"id": "train_036185", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "5c6b99fb0b525e553ec62fdcb902120462267cd7"}
{"id": "train_036186", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "978e8299753f655638ab0e465225e827cf895137"}
{"id": "train_036187", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["evaluation", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "3eb7c7dfee5e0f884c751abd56541c3bfaa0101f"}
{"id": "train_036188", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "06d06541c723200ed8bff4ce14ab075bccbf5ec3"}
{"id": "train_036189", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e4b7bab39961e61d29e2539b60367e9e07073fae"}
{"id": "train_036190", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "30e2c0eeb29d2806ff60be6325a3adaf8ddc260e"}
{"id": "train_036191", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e3b55eec44929cf23b48272537c05e7f9d105097"}
{"id": "train_036192", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "589ed5fc18568a53bdb684e7ae4667e355a8bec2"}
{"id": "train_036193", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "b7aee58fcf66cdb76216730558a3d6d28eacd546"}
{"id": "train_036194", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "169560cd9b0b0ce69f6193d4cc2b5a3098baed0e"}
{"id": "train_036195", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "445d9edee3bb15a6ffd6f1536038ca155a0097b2"}
{"id": "train_036196", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "148f9dc2b26ffa63a102fa5ff6ca4e5727a7b867"}
{"id": "train_036197", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ba26b2178d0d26441eb006eca2e216eb7a259f71"}
{"id": "train_036198", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6355595532d31a3e827732aa4bf76019365601cf"}
{"id": "train_036199", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d5e2639b36350843a802de4fe6371b53144deed2"}
{"id": "train_036200", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c05e6255d5e68628445752c741d5b1795c3f881d"}
{"id": "train_036201", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "63f5fc8e110c2651d97ad904b58c1d07aa32e574"}
{"id": "train_036202", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "120e37891f274fb6585647f2a38baebb06475de9"}
{"id": "train_036203", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "339262e17c493443fc8df1c20e82afa473532e86"}
{"id": "train_036204", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "268bbcff3fc7e7d43417528030467529d89a5832"}
{"id": "train_036205", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "1ebde0abf6784fcf3113e1805e2e7d73cda3f65f"}
{"id": "train_036206", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "5697004df0248bf345738b6a9aa294eebb41b156"}
{"id": "train_036207", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b9283878999ec42dcd8c864e4b847ec36942cff6"}
{"id": "train_036208", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2a0f09b4090a07a9f513a51d1230bb418b96a9e9"}
{"id": "train_036209", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2614397a5b2fb1c694108652d73cd2b33c864fe9"}
{"id": "train_036210", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "2a9d4835f9989093bc6ea0c82fb6c142003ff6d3"}
{"id": "train_036211", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "50b74975d083a71ee7804c9defdf1a75b02957f5"}
{"id": "train_036212", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "3e595c05440d7a300c9acfc5a75a42ed5788bc05"}
{"id": "train_036213", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "bdf6dba2cf9b0feb041fff864b1ddf415b9f12ab"}
{"id": "train_036214", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "3f6da69d11c9d427ec30a77c14fb2d554ccca49b"}
{"id": "train_036215", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "aa789f78a7a3a66251e618ee777dd652f2908cca"}
{"id": "train_036216", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "43e6989e3fd3135c0b3f85484051b0adce70d97d"}
{"id": "train_036217", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "5ab5d8ffe83b7f7bd954ab516e199e097b964584"}
{"id": "train_036218", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "0776f2e8a58ec06d09508ec2b82de253413ea959"}
{"id": "train_036219", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a42b138983a384ea4ef9327b4b687732eb2b9440"}
{"id": "train_036220", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c5c5d35bac9e812ba6e8e2ffa5788f11eb91bad5"}
{"id": "train_036221", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "109a04a727800c2635da8582f97e2c13128a5e5e"}
{"id": "train_036222", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "45d8975457e32eac9dcc9876fc6931341112ef98"}
{"id": "train_036223", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "565c9efeaf45f5b73bf391bb8562d52b569bbdea"}
{"id": "train_036224", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "7d8291ec27aeae1cdcade16b7765345069764438"}
{"id": "train_036225", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "757d938eb113915c294cfa97594d02747b0b5643"}
{"id": "train_036226", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d5fb5c9d208f692cfa60d25feb941ed1eb81a7c7"}
{"id": "train_036227", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d17e04da91d44b998ee96414f657aae01b596549"}
{"id": "train_036228", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c09e34ee3e56c21c4a66061200743b34e6cf3373"}
{"id": "train_036229", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0aa009ef7e95813aefc03ba07a4121ed1bdf261a"}
{"id": "train_036230", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "bce0bb1fa04719c743b967b6ab12b073d7fd5cba"}
{"id": "train_036231", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "4a72816028741978dc12e78c7fd355633ceada9c"}
{"id": "train_036232", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c19ac81cc2883138bd7890f46482f68dd352944f"}
{"id": "train_036233", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "4000ed3f632c4ffa7dd463de0828ceac2b29a361"}
{"id": "train_036234", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "0924f777eebd3033d6ae531e53ba9012e4a683f5"}
{"id": "train_036235", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f4861990ca274c58f0904bb0706d81402225c778"}
{"id": "train_036236", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "e55a4b580e3ba4dcd1775e6b514075d78bb6a727"}
{"id": "train_036237", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "8af3a97fa7750d6e7ace60a9d61fbb69d08011dc"}
{"id": "train_036238", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "8088ad2c6ac9daea741438c4915b042b79b212b9"}
{"id": "train_036239", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3a516dc5064803966a2d034f3c87f62578e98596"}
{"id": "train_036240", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "8b73e6de9fcf103aa30ed731da75dbc0256e3296"}
{"id": "train_036241", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "263277ddd83eb2ff96444c7d50ade9a8be5f9779"}
{"id": "train_036242", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d296e7e77079873d186204de0af718ce32131b70"}
{"id": "train_036243", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "9118002a731599c5ac6362d04a539c281b77aa49"}
{"id": "train_036244", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "84366419df0161df59a54493c1ab75165a55c743"}
{"id": "train_036245", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "af683eca50879e67c6ac8be6ae6578fda552d566"}
{"id": "train_036246", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "22c657ee649b24b27be29ca99b70c62f8a13ba83"}
{"id": "train_036247", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4ae776ad8e5b652da92f0b9ff69ecf961d4d92fd"}
{"id": "train_036248", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "993843a922ccf6152344df669ec67423e3c7db8f"}
{"id": "train_036249", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "13f1cd82e94333b0e2e85d589cdd81e24e426c9e"}
{"id": "train_036250", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "f64a705b01d1995a5dfb0bde206c119f71723fb0"}
{"id": "train_036251", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "694db0a9bb0a5ee6da216836b1863a72d9196508"}
{"id": "train_036252", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "049f3dfc70a089b7d2750147e3fb845e7b482d3c"}
{"id": "train_036253", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "ee265008b4c8057ede4cc4f39777810105769603"}
{"id": "train_036254", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "8f8bfeddcfcd6b0c62f20940466059d15b587945"}
{"id": "train_036255", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "f29a82a1a19a8de2c59ce16f50bf1f5dba19cb6e"}
{"id": "train_036256", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "7e3d6ac9c4f7b2c729aca04805cbb5b5e700068c"}
{"id": "train_036257", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "822b51e748e41016ca12f05d9e182553579ceed5"}
{"id": "train_036258", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6b39e468638ec98504a51b082399fcc1910c26fd"}
{"id": "train_036259", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4c595f2addc8a99a6bae2bfb7e55a449e175b916"}
{"id": "train_036260", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "049bc3ad2fe5bed8f10acc973603dda6f57b80f6"}
{"id": "train_036261", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "88b84cae9765369fc51b273aec0a29cdd61fdadc"}
{"id": "train_036262", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "097d2ea0310b86ed0443f7a72db6b840329238b0"}
{"id": "train_036263", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "da34f0258e580105ee9f3bc74824dcf2614b37b6"}
{"id": "train_036264", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3b98f9e012d8cdb72e9689203084434452009114"}
{"id": "train_036265", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "aea05c7a1ef3a2a72daaeedf7f0ad65c697bfbf9"}
{"id": "train_036266", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f471955f0fa3edb2ec8d21fa57618cd21f4ddb42"}
{"id": "train_036267", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["security", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5d7b89f77ff7a3fc0f1fe604c7806dbd37b7c4b4"}
{"id": "train_036268", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "455ae1258076b85a95c9c704abf3aa72ecd96cb7"}
{"id": "train_036269", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "733d53baba355d96187746fc4901949dce111cdf"}
{"id": "train_036270", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "5587280b315ba3a53d220f95a4ce7da21d9143fb"}
{"id": "train_036271", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6e91c974b0ab75ded584bef5c1dac9dbcd867bcc"}
{"id": "train_036272", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "268cd4efdfb4fbb7b9984744757e3836b68815dc"}
{"id": "train_036273", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "610354886f4c10647caeaad0c5421edd629ee90d"}
{"id": "train_036274", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d1f29069b3fba5e06fd9627e97da3681492f460b"}
{"id": "train_036275", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "42e60aa46767e6722eb85886eed2870b6ca191e7"}
{"id": "train_036276", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "67feef66bf10d67d2e3595d255812cec32069f6a"}
{"id": "train_036277", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f3a5e966661b18bce6b8d0b0b36b1de2b2113904"}
{"id": "train_036278", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d783ab821be2019d5a7449dbfeb35245b4adaa04"}
{"id": "train_036279", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "9204b3f7b80b4b1a37df740f3ad7f9970413d8a1"}
{"id": "train_036280", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "902618f2dd6899315a75e6be5774a447334d85be"}
{"id": "train_036281", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "eb8e1ca64a673df36b22d8af0b882cb622851663"}
{"id": "train_036282", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "34e1f8a9cbd01dba445fa9633998d7bc41fcbd9e"}
{"id": "train_036283", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "30f9e4e4b0cffea1616fd2085b11678606e8ea38"}
{"id": "train_036284", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "9b35bb29f9c74d8148dbae10f52f8df4a934b72a"}
{"id": "train_036285", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "cbc429a544baed70a62a8659fd7beb838f88785c"}
{"id": "train_036286", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "7812edfd18e6ad5407bdadb7fd7e3d6a7950817f"}
{"id": "train_036287", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "fac0454c20e3d4e9a6cf64a55b5b8b728361ca70"}
{"id": "train_036288", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "1a18e489e761f82ad2a562dc25d99fae80cfa961"}
{"id": "train_036289", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "f060a96a7e0d4f504f93f07da5c7239379675e3d"}
{"id": "train_036290", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "8cd2faee073faa53efe1129d20e4301a92f96a00"}
{"id": "train_036291", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "d89436799e3c84497b8f8c1081f059d8ff8e96fb"}
{"id": "train_036292", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "4815f772a04ae36eec9cd9a7aca087d0f58a0082"}
{"id": "train_036293", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "dff77b95065a42aa7d55308307829beec110f828"}
{"id": "train_036294", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "20872573475dc6041dea1cb49636fda9bae4dcb4"}
{"id": "train_036295", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "b67a5288dd8dbb69f55ecd68c790607b62ea1a5c"}
{"id": "train_036296", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "5e494c2e561f2de3fef14839d0aa12dbe358f361"}
{"id": "train_036297", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "331da0e84c6489e58a9392f4debdbb2c1ea78105"}
{"id": "train_036298", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "26005f11970a73d3cc939737afa613991f9b7513"}
{"id": "train_036299", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "79f985e3f3aca6ed482a17d1e0c4852bcabc9068"}
{"id": "train_036300", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "6a513619cfb309801b281d64ac7e71efef065698"}
{"id": "train_036301", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "033f6c0633998d093132fef039289aafa35bd3c5"}
{"id": "train_036302", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "5a6a3f27d786629f31694e2f9f97237fa51e8a87"}
{"id": "train_036303", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "3a05de53c8b1a24844405f0599fbe0a542dc1bad"}
{"id": "train_036304", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "0f023b9ea076ef899af8b5a47ae947a2122e6c78"}
{"id": "train_036305", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a321ff685d63fbaf813550c6fe0057d63b3dccc5"}
{"id": "train_036306", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "903618e154625e050c6f30830cac62205f5ddbcd"}
{"id": "train_036307", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f53972549c1445e30ffe336dfeccd4c2dad39a9a"}
{"id": "train_036308", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f2b6170acb686c791a571573af9b373b118ea091"}
{"id": "train_036309", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "fa947277d0603b74a1c127a5598d674882ffe7e7"}
{"id": "train_036310", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "5beb90b88924699e78bc07b13299854eaaba0934"}
{"id": "train_036311", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "94f9c9640a499a68e34ac18f54162c7dcd8663dc"}
{"id": "train_036312", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "56066dc3f132243c5b291216d7cc54f8145766fa"}
{"id": "train_036313", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "147794813aae1471bb05ee50876e8783fb6d3dd8"}
{"id": "train_036314", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "0e096a8c8df7991b6325f9e84e25c037f5a274d1"}
{"id": "train_036315", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "ed3ad4063795f58620907d90ee72f2a53a8a6be6"}
{"id": "train_036316", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "027400c74f50707388a83706bfb7c3840d123f8d"}
{"id": "train_036317", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "1a2f11cf614b2487b30737dcda03793ebf418181"}
{"id": "train_036318", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b9405b0805e54ed1c719387b1b6d3b0fbf097803"}
{"id": "train_036319", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e0ff3a7a273ceeb883099a9701245cedfaad9176"}
{"id": "train_036320", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "10afb4c993b60ccf7562554e609b669d7bb93d2d"}
{"id": "train_036321", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "a1e3c2aa60d1e49eb2f1c07d2007fa0240a5cfcb"}
{"id": "train_036322", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "89080205ba16bd4bf40664c01bc498d2a0dc1353"}
{"id": "train_036323", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ad1e0ce67f3e222d111bdc53d2439738066e3209"}
{"id": "train_036324", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "61bea16932c7c5b6cbcbe258971b0488909c6492"}
{"id": "train_036325", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "cfc7a750c4c753acea3231304368b9ab2662e380"}
{"id": "train_036326", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "59aa170e106b077416b118b92a9d59a54314522e"}
{"id": "train_036327", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f4622bf8fb7c7f5452a2bca2c8d61bf787bf45bb"}
{"id": "train_036328", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "273f458ace3960f2ec5254de4fd8b0902c12be19"}
{"id": "train_036329", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "57ef51cb2d7e1ad9e0e6d2975e4ff1646f7206af"}
{"id": "train_036330", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "94f4096de9f24cb0a2a7afa77c00eb4c6bc04bfd"}
{"id": "train_036331", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "25d3678a2eb725c2d1460164b6d399078e5551dc"}
{"id": "train_036332", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "eb109b535e5356db691f54d5564f7ebe5bf37d58"}
{"id": "train_036333", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "75ce1f13dacf3d1f7314cbbe4f2b588e0346af66"}
{"id": "train_036334", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "9179d007fad0cd04556bf5cb51b3848f045c3308"}
{"id": "train_036335", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c7041f01c74f1d6fd6f260d16be32b5b4f1a0eb8"}
{"id": "train_036336", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "7b63c669cae428cb8c9b9f1f8d8563680cf14ac2"}
{"id": "train_036337", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f2bed33c5604472090bdc09764917972d9ed6541"}
{"id": "train_036338", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "fbb5fb7126428184269ff9a02372356fa842ffea"}
{"id": "train_036339", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "bdf6dba2cf9b0feb041fff864b1ddf415b9f12ab"}
{"id": "train_036340", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "65620079a7a1dbce8f16975e68b6c3356dd4f672"}
{"id": "train_036341", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "25d6353adc35801f4992e43bb3a35ff10fa51922"}
{"id": "train_036342", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "22980b0fdfd5efe75a15f2a1f88c0cf7ed0eaf94"}
{"id": "train_036343", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d994d886d4b7bcd2dddd112d5de118bc707e3d1e"}
{"id": "train_036344", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6381b135475d359e89fb72d56a60ca45d0663fd6"}
{"id": "train_036345", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d721870d8b256604a55aea6250f4d82af59d9a72"}
{"id": "train_036346", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "40d36568cd5b823bad35b76eef7d9af7b3b94ea8"}
{"id": "train_036347", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "ccb9cd391fb3a4737412fefb49803069ce081894"}
{"id": "train_036348", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "237b30ce8272f5c13df8b2b18d70b5de5ece1b1c"}
{"id": "train_036349", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "0a7aa09f9c75f45732f57808c5fce66c79191e7d"}
{"id": "train_036350", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "247df3b2d7c608d3230ec693d88102a1b8ad7a04"}
{"id": "train_036351", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "541fa9c4ed8883b82cf30a88976ba6d8f4035ca2"}
{"id": "train_036352", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "674f2f6cb1d8cc7f77c67a1941f9929e05b66f46"}
{"id": "train_036353", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "710479b385359edd4a1140097b03cb115ba77551"}
{"id": "train_036354", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5451324b514c92101126595dafac1c7ba50ef8fe"}
{"id": "train_036355", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "70ad2e10ca304988d446bda2cfaafa12f9536ffa"}
{"id": "train_036356", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "16c4a34e95b33e21d87bb60cb5b8fa0adbccf7cc"}
{"id": "train_036357", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "0830035c8ddaa7a3ee5baafaebbd49bcfdee2166"}
{"id": "train_036358", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "77b473a89daeaf0715f6262cd12af5cd3aefba70"}
{"id": "train_036359", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a18ef852f425fcad6a654f11f01890fb471d9c25"}
{"id": "train_036360", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "f7f16d436beeb3f51dfb96b425bd0e1ac5685515"}
{"id": "train_036361", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "c547ff3494a701f38f9cb9bd3e75db9e7cb09230"}
{"id": "train_036362", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "3f41ce15e096bb712b8ca9e0fbc66ff619b5e27f"}
{"id": "train_036363", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "e526acf33f41bee10a1e87306ea81e5b6760edef"}
{"id": "train_036364", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d9e031d07ff0102d0acc16bb403ec714378a7b70"}
{"id": "train_036365", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "8dd090b68cb3bc95da4819d50989d0e40956e2c3"}
{"id": "train_036366", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "321b2e5cdd2cfd47828bd70c1684f342871b049f"}
{"id": "train_036367", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "edfdfbff4a7977f5ffbd5e340a3ef6128b06b218"}
{"id": "train_036368", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "58e68b4f04894799acb395f27c4455d11178b9ba"}
{"id": "train_036369", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "b1df22b2c812a40fb00906cb99f9a90690e338c5"}
{"id": "train_036370", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "75b622de54d3c91cf3cd5389cbf0855b3c470605"}
{"id": "train_036371", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "215d53c301145bbb5e8bc95b0ef449048f068e7f"}
{"id": "train_036372", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "8b1e89d1eb4a57baf5a56903d58270a461a13348"}
{"id": "train_036373", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "badf5e120b9a1beb2cd5b3f7c9085182b31c2cb9"}
{"id": "train_036374", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "182655eeaecf38f431b6db425568f46031e5a2d0"}
{"id": "train_036375", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a872f7b2c2091f1c226602bd0383f7364ee9616a"}
{"id": "train_036376", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "86ca7dab7f6f5304c8b0029f27d9d3c57c315e7b"}
{"id": "train_036377", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c9ea87c56df24ba6d83436c5a71623942d826c81"}
{"id": "train_036378", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "35cd2fa791fb3a2b46a3bd9bdac6c9fb4c9825b7"}
{"id": "train_036379", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "df670df6c01b2143ab4a5fc82f5ee874146b8706"}
{"id": "train_036380", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "536fd7059724e50a426983a330310fc70a9f9ec9"}
{"id": "train_036381", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "65db80439e85fc7c113e93c8a1f3e02ea11b661b"}
{"id": "train_036382", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "85b397c2642b0169ad9600feb9722885993678ba"}
{"id": "train_036383", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "4ce6d79f1233c2e123adef635731e17ad28b8300"}
{"id": "train_036384", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "c06d574fa2457f35d02246f677928d21356d0a2e"}
{"id": "train_036385", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "d40a848a3005d98a0a8bc38e046454ede1ebe4dd"}
{"id": "train_036386", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "ed6746ddaa8799893aa0ecc9322f0e062330333a"}
{"id": "train_036387", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "0ac85336095dbd141f9afddf3f16054784799e35"}
{"id": "train_036388", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "e4ce4e4974826c61c5c570d5256910d0369d1ee4"}
{"id": "train_036389", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f1a96f39ba88c1934fe8864528a0fa1f38aa6aaf"}
{"id": "train_036390", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "1b97eb963f1e8ee0072b68a8d1d765721804590e"}
{"id": "train_036391", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b067cedcee70b4f80e8b1409bc6ab6f110f846fe"}
{"id": "train_036392", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "351185c7bb13413282ff375936b3a1ea00b8abca"}
{"id": "train_036393", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "02246aeda5e2de81a1b266c71c22c595f3597841"}
{"id": "train_036394", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "42168fc60098e4157b3433646f192521d706e916"}
{"id": "train_036395", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "1ec5f20b3605f9e6bafeb966a8ed45897c04d680"}
{"id": "train_036396", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "3f3263ed2e7781ee1f22c3423944c956c8033dd0"}
{"id": "train_036397", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d279e8aa68ea7b585b9b7839fac97eab059cb9f2"}
{"id": "train_036398", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "4c2d7e98302e1a276b7ab725b2627481d2b6f1be"}
{"id": "train_036399", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f62f54e96c37d1365f3d434396ccd21e9c2b0cda"}
{"id": "train_036400", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c6d25c39f94742e5576ed0093f6d235681b4780c"}
{"id": "train_036401", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "a6a5630c76c917bd7248a4c92a5b11b0396e4459"}
{"id": "train_036402", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "21650e0e9a89e6eb40fdbecaf9c7f88de262e087"}
{"id": "train_036403", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "80a2a1ea1bffae1d136ecc9a52b76cab9a56ea91"}
{"id": "train_036404", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "deaf482e0fa389bc8e7cef8e03c3a9798304f458"}
{"id": "train_036405", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "319bce6183eb198233d06b8bb3e3d324dd4cf3c8"}
{"id": "train_036406", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d8b858fcade573d9bbb63a963dbfc78b86427b52"}
{"id": "train_036407", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "5c6b99fb0b525e553ec62fdcb902120462267cd7"}
{"id": "train_036408", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "545eba322603c9e5cf50876574a46453c6f1d68d"}
{"id": "train_036409", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f501596bcb5f18d76c131cd4212e39c1abb15b84"}
{"id": "train_036410", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f79618bb0a205e5badf4a2840791d040afd0861e"}
{"id": "train_036411", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "f5ba595ba17a588d39017f76f7619e01ca760f75"}
{"id": "train_036412", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "55ab1ee0a33413a0e8fc6829c377872e6b9e00f4"}
{"id": "train_036413", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "314af58bfc72e0e29fad60a0f3917839e05731c3"}
{"id": "train_036414", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "90ccd49a2bd900e9ad8757afb07fb99b50d4364c"}
{"id": "train_036415", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "75c303c76575a7a14825be4151aa7372bd078e80"}
{"id": "train_036416", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a5f881287b05bdcd283c8b7b800ae4d39b6b9c51"}
{"id": "train_036417", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "2e1364d0e154059f86025bb78a79692560e6503c"}
{"id": "train_036418", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "889b9753d9b25c6ac9c6930fb104713d5715bf2a"}
{"id": "train_036419", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b1f54ef5d390a347ae746df4f7b9c8c762c0b418"}
{"id": "train_036420", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "4f55f4d876656593445686ecca258edab0a08817"}
{"id": "train_036421", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d86c95ff8cac12b5831cc88ce61d80637aab1cb7"}
{"id": "train_036422", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "8a424b249c4f5b010541a7be37adc80ea3668f53"}
{"id": "train_036423", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "05a439d9529abda000c3d9c7a4edfa3d8b59ddbf"}
{"id": "train_036424", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d9d7b4ea5a4ba5eca2f5e6fb00f4c332515dc050"}
{"id": "train_036425", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c26fab5a069aaecda2fbf5ac5e97e164fa54dc36"}
{"id": "train_036426", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f704343cadd41a848996848e0f25320d6ffb4468"}
{"id": "train_036427", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "9ee1be780ae7ce43a7c19007fb57f969f105ede4"}
{"id": "train_036428", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4d19544c57d92d7cae2989f1b2aaa3a1a286b62f"}
{"id": "train_036429", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7a7782c7c07d61d600e7b082cf93547a621090dd"}
{"id": "train_036430", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c398013de3dfb709c6f51531f37a9e9a84f1a2e2"}
{"id": "train_036431", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "afbb44e24c93778d7b5ec566707d614d850a100c"}
{"id": "train_036432", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "0ec0f6b82af6153cc096e979b1add1f95aeaa191"}
{"id": "train_036433", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "943aaba7a371c1fdb692fb087bde346f888145ac"}
{"id": "train_036434", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "a360514940f404968d4f8d7614fa44f58764b4b3"}
{"id": "train_036435", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "33a9bc1160df62e000f9577a05fdeaf19be57f42"}
{"id": "train_036436", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "26dbafae46cf979bf1d27fed5033c9a7b0dc8e7b"}
{"id": "train_036437", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "6d82a018a67ea28970046b2532e6b8ab7b69de06"}
{"id": "train_036438", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "3113e5d80445c581af4cb9c85bb53de49e13bc41"}
{"id": "train_036439", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "072ec5350469d505b4c50a8393065af766cf783b"}
{"id": "train_036440", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "ae1875a74e9c2cd933835e6a793801d3dc17412a"}
{"id": "train_036441", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "6171de312b3f5238ee8fa70fdfff49c160ac5beb"}
{"id": "train_036442", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a54a43d56de2fe9bfff0f554d27c776444bfec6f"}
{"id": "train_036443", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "f6418197f72cb026be1264e827eda679718138bb"}
{"id": "train_036444", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "12f0601f80ae3d2713b6dea7e627d43548c5a112"}
{"id": "train_036445", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "93f92623356fdadb506f61c1f95de95037827807"}
{"id": "train_036446", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2637586b8386e4a44f52525770148f5cff0b14a5"}
{"id": "train_036447", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "63b2dc2218ee592f1cdd2f99de52862b621144df"}
{"id": "train_036448", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c13e125a66350ce85cc7b44b6158b11350f5912e"}
{"id": "train_036449", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "61d8e4f042e6b467bb6249a155d9d3017919e6cf"}
{"id": "train_036450", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "81ab9745f764b1e3979fca5fa63a178be84d1aff"}
{"id": "train_036451", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4d725e4ea8440e3de73f9e8a3550b7802c68bc35"}
{"id": "train_036452", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "a566fb8a39d1f1624818fd322ba419dfdb70689b"}
{"id": "train_036453", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "3fb57f9f27a01dac34d24e745c6cba72710d3169"}
{"id": "train_036454", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "54693c17a3193df6b7f7e061c645028afba19562"}
{"id": "train_036455", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "ae5cd7b9bb398c1a59138fe7b2c5fec737b66196"}
{"id": "train_036456", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "9bf4d79f94061e4e784061bd386d298ec2e80802"}
{"id": "train_036457", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "b2fbf978e8c01760f6dd0946bf1cbd7a5be9bb2c"}
{"id": "train_036458", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "4a773d76b67466d487f0590c32212a2f834bde68"}
{"id": "train_036459", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "168b97b4851f03804aeff67bb1c47500ed537c72"}
{"id": "train_036460", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "a71f6d21ad14e70e5c442833c8b9ab4bbd1f6067"}
{"id": "train_036461", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d4fc64098ed8a9e6c1c4293d4bb4341f47146899"}
{"id": "train_036462", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f2983b3902b1c13871da277f75cba40f14943d52"}
{"id": "train_036463", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "cde792727c9fac764c602170542497ce5b1a70c5"}
{"id": "train_036464", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "083d8f0662b2468da0a3dfbeca43663a880f4352"}
{"id": "train_036465", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "bd0592448b31ddef29e8d764c91c687d782c5099"}
{"id": "train_036466", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "5af898491fbc9688545c91d9b054e66fe4a90c0c"}
{"id": "train_036467", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "713e5e9e5383d9f72e8245295a4afc00b5dd42af"}
{"id": "train_036468", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "85db00c63fffa6f80f72fb1fd5652a5752dcd4c9"}
{"id": "train_036469", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "7eac6b4c117ba4b3d6163547ecd115ba4a7318fd"}
{"id": "train_036470", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c098529ca99bfcf6fb164b3209d0d8879e15ba95"}
{"id": "train_036471", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "603e20172a90dc8cf5ed5ec5f6167beaf1ba2ea5"}
{"id": "train_036472", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "1e82775137d0900af6dbd755db5069a35f2a08ca"}
{"id": "train_036473", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "bcbd3aa62af828dd086d48baaa530fb6933ceea1"}
{"id": "train_036474", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "379f7701cefb587053d317f92f7732b098d024fd"}
{"id": "train_036475", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "28ace307046eca4327778dbe6ec29a4eb600e25b"}
{"id": "train_036476", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "fa8698f2e37c3a6145dc7834630bd8dc35a7ece9"}
{"id": "train_036477", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7af3afc51cae0b4ffb1dd194b6879bfef118cc18"}
{"id": "train_036478", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "8f261c80109e98f0a89799f189fea8ab0581a22d"}
{"id": "train_036479", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "379f7701cefb587053d317f92f7732b098d024fd"}
{"id": "train_036480", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "b5e3a5215bbdce1f39b10aac73770eda9d62bfd7"}
{"id": "train_036481", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0b59ec71592723259a9c69728ba5b2790617fc59"}
{"id": "train_036482", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "06459876234ce5458f3aa1cc9d1eda4b7e120161"}
{"id": "train_036483", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "a8304293f0689e52e8ef15efb1754124be992c17"}
{"id": "train_036484", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5aaf5d6f63e739466de648b21b6b9b078fdabcf3"}
{"id": "train_036485", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "9b362388f5ad1eda238f17a2a993c07def0f3aed"}
{"id": "train_036486", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "5aee0539737f8cb5f2f48d9ba249e033d7150817"}
{"id": "train_036487", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b96da8113eb39868fa729c959a860a44afedbaf2"}
{"id": "train_036488", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "080a1350ad95ce284931f076cb0d860300090459"}
{"id": "train_036489", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0bc8f65b8f6d3a42e7c12a21ec619503c2666330"}
{"id": "train_036490", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2fbf696cc33439cb49c219586fdd760bac37d9f7"}
{"id": "train_036491", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "708716251ae5898b7547ab9e54ad4ee18cd03cc2"}
{"id": "train_036492", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "42e60aa46767e6722eb85886eed2870b6ca191e7"}
{"id": "train_036493", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "2932f612a5a364b603ff87b1be3ba7975d7eb5c7"}
{"id": "train_036494", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "cc368b0c38d4cd760bef2838b23e53c7d5acaa1a"}
{"id": "train_036495", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "5800933bf77c37824068a2902675778c7353071e"}
{"id": "train_036496", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "111991b90f705f2e1904852e4b2e3a70229d099c"}
{"id": "train_036497", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "5587280b315ba3a53d220f95a4ce7da21d9143fb"}
{"id": "train_036498", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "380293d614ab443e2dd518c4eb16a4adf66783b3"}
{"id": "train_036499", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "f8c0930a9e2eb49f62a1ef61bb69fba9d0836211"}
{"id": "train_036500", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "21874300007dfce615ebe8a31e18ea1521173a64"}
{"id": "train_036501", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "858a45dc996ddfe067ce3d3a19fc748c013a40f1"}
{"id": "train_036502", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "fc3a96dffe1a961c4404818d720eab9020fe0280"}
{"id": "train_036503", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "82567ede3948a6fad57a512f34e619d5b15e6211"}
{"id": "train_036504", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "54b81a2c06d878f70f79dc6238f9854fc0e5a735"}
{"id": "train_036505", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "5919ddfb6d50c00b003cfbf75c5dbe323e7be8d9"}
{"id": "train_036506", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a2fbbce6a4e7413d86337f57aec831f1c5bc80b1"}
{"id": "train_036507", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f31161597fc61b0002e1cf438263862d8b375be8"}
{"id": "train_036508", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "0bad4323a0af4e0ddef61538cd48e7031ca8f76c"}
{"id": "train_036509", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a2d2c6afe61f2a2e4fd39a13eaaf3556d63ddfa3"}
{"id": "train_036510", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c4440c5096ae8b7d93ade5d35955f9865f456362"}
{"id": "train_036511", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f0abcccf022c58d228c1b0980f205752d33d3d8d"}
{"id": "train_036512", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "16357bc1627084dce4f4151af9fbfebc0c68156f"}
{"id": "train_036513", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["data", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "273261a4ce4511d40a7d0ca3879bb1d9d1eb82a4"}
{"id": "train_036514", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "b28744a4c20fe7a8aae3a1752456827ec0acc1b8"}
{"id": "train_036515", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "00f27b4a2fae26124e42bbfb5bca315de2570638"}
{"id": "train_036516", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c3295256f367cdd8f15ba93dea199c6cd675b34f"}
{"id": "train_036517", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "97d9dca1088e97a7dcc7714de3f3048a0300f0d5"}
{"id": "train_036518", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "46ccc56a958374dfc44472ff61aa09a32b432a39"}
{"id": "train_036519", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c1079a4aa7909fa823a2275ec6369c8425ccdd6c"}
{"id": "train_036520", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b22e00424fa1dd4a49ed2e76ca3c883b930415bc"}
{"id": "train_036521", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "318e8b8ea6c125b532235481460ebd92296a2108"}
{"id": "train_036522", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "2076406f7fd2e6f3ca5496920d5121d3e3a9e1e4"}
{"id": "train_036523", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "c93b2b3db404018f8fc5fe2fd33173d7ab608933"}
{"id": "train_036524", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "fbc821d3a78928c8281b05ac25a0131bd4b3513d"}
{"id": "train_036525", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "2182d8c4d140d33434833b9b453af3c4df06a36d"}
{"id": "train_036526", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "cc3a9388eb540f2592024697a895f68c6e4cdbb7"}
{"id": "train_036527", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b6d8f6232b2adb69145efa29843a24ff7dfebc17"}
{"id": "train_036528", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "2bc3cf10648a9df1fa3b36f65b502fe59161431b"}
{"id": "train_036529", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "4c2f1003f0e0d351b9231d58d1b8f469396f7967"}
{"id": "train_036530", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "7c37e065d5c25f70f685d7601795e27ea8bc9cbf"}
{"id": "train_036531", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "cb72b2c5d24228e720ce6f1e75d1c19df67fd683"}
{"id": "train_036532", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d8c656038ff0fa1cbcd057f7bf99cfe60e0f4057"}
{"id": "train_036533", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "009c86852e0325875b9ca8cd17c849e540160c7f"}
{"id": "train_036534", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "ce6caf0ad86798d0e58db653d1f4e21d96801f90"}
{"id": "train_036535", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e15141a378d2ea89f4dbb2a65e783132187bc9aa"}
{"id": "train_036536", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d64efabe726093192460ae46f0411fec82f73cb1"}
{"id": "train_036537", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "602114878339931632124edddd1f13615afcd4ee"}
{"id": "train_036538", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "4adfb0c1ceca580e53e4f403b91577854d87c7fe"}
{"id": "train_036539", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "c5066c32a774d009605cf605dc241a9abdf95520"}
{"id": "train_036540", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "eeb1891fd74cf644d052875aee999eb68024fa60"}
{"id": "train_036541", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "b6f5615fafbf201c532e2e165a57e08c8adb8cf4"}
{"id": "train_036542", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "8a33e429b0817c7bda9d3061aca23c8668a6db40"}
{"id": "train_036543", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "0fe37ba9c09a0a8059494f877f395ada7e478dee"}
{"id": "train_036544", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "6d48f9391d5823d521b5d254cb72b0f2e55b884d"}
{"id": "train_036545", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a59696228bbf22f740dd4f550c85817fb4c5564d"}
{"id": "train_036546", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "c7afa8c3847be59e40c9ccad00785eb80ad4078b"}
{"id": "train_036547", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "492faa2de733c1b3f675477adcc1e6e526f3a4b6"}
{"id": "train_036548", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "e92e895e56a9e5f3c0c944c1b35dad7175ea630f"}
{"id": "train_036549", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "cacea8cb7f073d082c0c6e4b9c12cec445cdd202"}
{"id": "train_036550", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "e0d9f5d742b75cc4e852c67f8a890cbd714a5a56"}
{"id": "train_036551", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "daffe6544cd5b9d65b815e9cfd1a997410b57289"}
{"id": "train_036552", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "bd1975b63fdd356adeecca2961c1fb26da694c97"}
{"id": "train_036553", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "8dea89501e07ee3a81fcf324dd03ceb70b526dcd"}
{"id": "train_036554", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "53f22e47271dce66d7cc697705d1adfc8f1b90b8"}
{"id": "train_036555", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "e31a721348c312c04ac2b9c77a62a8cac1ec38d0"}
{"id": "train_036556", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c05791bbc926e75c752f5436be861334f70ce93a"}
{"id": "train_036557", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["agent", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "dae5993a4c7302091f7f280b33e18d31adafe8f8"}
{"id": "train_036558", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "bdf0414ec167a3afff51a645c6c8b5806a37cd6f"}
{"id": "train_036559", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "144379d37767c9b0c8b609782e359489f634b7ce"}
{"id": "train_036560", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "cff9c5e6cf36a0318ffa887000bc60807765b7c6"}
{"id": "train_036561", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "9ae48f99e18b3f1ea5627075effecb2e0cf009e9"}
{"id": "train_036562", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a6c8ac3f2dc1ebd09c2a7af8a11d17972e9b1aa3"}
{"id": "train_036563", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "5ca7a2112da029e3f0d3c09cded36c56604d7c75"}
{"id": "train_036564", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b5c19ab57afe9a074c4c773c6bdd267ed5f96811"}
{"id": "train_036565", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "cdaddb9931dc6581a22fbf1267be3099d1224630"}
{"id": "train_036566", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "7bf869c7a5cc71e6a6829058c9fd1a29623205fc"}
{"id": "train_036567", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "2c844f109a72e96dd4d00725702893a7b6a1666f"}
{"id": "train_036568", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "9a561e4412cd8548cee9394248e1a86fec2e451a"}
{"id": "train_036569", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a962c80a73eed5aff7a0f235b1976a60814abd2d"}
{"id": "train_036570", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "31e054cb2661f654fd7edcae423967043968bc8f"}
{"id": "train_036571", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "47278affa44a869f9af494ba7ee2a15241d01840"}
{"id": "train_036572", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "24fe50de8a333abe903220ffb16cd860778e586b"}
{"id": "train_036573", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "4af3701aa3a208bd5e7349876ec93ae203d32c6f"}
{"id": "train_036574", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "30559205ff1171096a4aff0e98ce1433aa1d20a0"}
{"id": "train_036575", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "92a69375df9aa8c64348537491a576dcd36a36dd"}
{"id": "train_036576", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "9ce2d0ee5ae1c382389b5f741267791d1b31b7b7"}
{"id": "train_036577", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "e299ec84837330835f1af5ba3e69a070fb2ea6c3"}
{"id": "train_036578", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "8f7a6aae00837b2913445c41dfa4dc7df1619217"}
{"id": "train_036579", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e4d4fca548d0d78fa4c130d4bde2faf30fb92755"}
{"id": "train_036580", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6c4d25ab7f699dfa166ca8c03ae5016ee14bb4ea"}
{"id": "train_036581", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d02830cd879bf6812cf6afb8261bf17d026a1421"}
{"id": "train_036582", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "6ba92939dff262160946801a3625da6360df99fc"}
{"id": "train_036583", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "7dbc5172a399311262a718319df63e718ca5b4da"}
{"id": "train_036584", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "2815037c0dbf49de6fb76877955b45c408455e8c"}
{"id": "train_036585", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "3149f56b2ab271a091a2d66338f317a9b50362c2"}
{"id": "train_036586", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "01cde61657faa2cbad5f9f2c0a7e6c8f5ea6735d"}
{"id": "train_036587", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "1c0bb18311f6fd6e6399496b77b79e8f151077a8"}
{"id": "train_036588", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2dc9388e3b9ecab34bc12f0758ab172bef77687b"}
{"id": "train_036589", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "7a26de52d4ae1c6892a2ee421e10fc69bd99ef24"}
{"id": "train_036590", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "eca7f9f546402669909688bace8deb3eb599176c"}
{"id": "train_036591", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "a5628cd15897b7428a3f07c46de3a837efc1723e"}
{"id": "train_036592", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ec2a2f1f3c6efe84b6476822bdc90fec44350a0c"}
{"id": "train_036593", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "b9e72b0bc1e9123906a113036dca77502dff0b7e"}
{"id": "train_036594", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "3aa6475f611ec4f8628d598d6789f8986546559a"}
{"id": "train_036595", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d7bd4283ba388aeeacc2a25f943232c65aad1652"}
{"id": "train_036596", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "dc10cf2ec3c5aa13a4addd4083f023dc93db1dce"}
{"id": "train_036597", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "bdf19d1820933577a151d18760d5c454b638c181"}
{"id": "train_036598", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "1db80b992d72f70ad76eed57b28688992c114c67"}
{"id": "train_036599", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d4e7f1865afa57f4f68c7803a6ac3bf0d1134e5a"}
{"id": "train_036600", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "0ba2b3b4c444b2d35681257f9fcae49b860f60c6"}
{"id": "train_036601", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d2854a1760b5b39ef174104153e9574d7f68ce90"}
{"id": "train_036602", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "33c4e6a1e0f5da92192d9b1a1ee81d45a191f8a9"}
{"id": "train_036603", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "60ba8ea401d0bf0e06c134e0a035890c102b97f5"}
{"id": "train_036604", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "05c85bc47d1c59df89780f6766a62e2d08e1abce"}
{"id": "train_036605", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "97850ce13fa0a9b29b21a8c328eef72b1b192dc0"}
{"id": "train_036606", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "fb41125ff1c750bcdaac8d542a050d35ef384286"}
{"id": "train_036607", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "106219da6843a4610641fe7dfce7d62e341e9b13"}
{"id": "train_036608", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e4855f4dba020c5cc0a6400fbccf6b5ad6318456"}
{"id": "train_036609", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2b3507467e7dad6fdc3162874f65a0c1dae9e823"}
{"id": "train_036610", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "32d76c9b44e5781a70c072f57c489fb4cc454af3"}
{"id": "train_036611", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e9aa082e20e7f7c286d92ce5bc9c21ed6e3e2a37"}
{"id": "train_036612", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "06589a2c9ea8475873e5db4c2c6258b310b68087"}
{"id": "train_036613", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4966d72baadbf14bf508beeb418ee0bc9c8fb08e"}
{"id": "train_036614", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "2ab17f50df5ee830d501d849dd9d3f6f43c3a24d"}
{"id": "train_036615", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "80d2cd8d2eab669084b7dac416de1f42780c475c"}
{"id": "train_036616", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "e6c88962efdc6072f3f9eaed9f710a23a968e9c2"}
{"id": "train_036617", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "383232752d6290c07409b92ae84c532fe1abed38"}
{"id": "train_036618", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7db9244fd291eec9072e4b36f4cf39752a9742e4"}
{"id": "train_036619", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "23b328041ee838622a9379d41a1cca9b4a7a8ccf"}
{"id": "train_036620", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f007f2995b7ceac2f73016122d09f279ab015d51"}
{"id": "train_036621", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b596ff9b0af62fd2d8c5bb3744f97b672c553ee4"}
{"id": "train_036622", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "3be6b7b857d3b481fec2406b3346dd888392af16"}
{"id": "train_036623", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "db1ddc4708db4aa52a4ed8dde66deb822f4e82b0"}
{"id": "train_036624", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "dea6f93a71c3ab79357fab396e4629b94ab2df38"}
{"id": "train_036625", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0302afcbca45d4961b34a940726c0e8f52ae1472"}
{"id": "train_036626", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "405c03de446997c76b81d3f0aab89e4d371e5403"}
{"id": "train_036627", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "64ac137bb70e89054f83097cd417aa1f4bdd7f3b"}
{"id": "train_036628", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "9abcb17007c37d903fecfe817a50d8376deff579"}
{"id": "train_036629", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "cf6b5e5fd351024d300e3aadd1b7df43230f0065"}
{"id": "train_036630", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["research", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "2a55fa20ab377a986b29071ce18703127f04baf5"}
{"id": "train_036631", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "85c490cf7c3f0e5c326bec8f56b618fe526a6966"}
{"id": "train_036632", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "3a018f89b0aa897ee34b07ad7c299bae84d9feed"}
{"id": "train_036633", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a3c7ab82662e2b5f9d1bfe20755e545454ab9214"}
{"id": "train_036634", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d784f9f8f448c3df2204c3340e0be1167e5e84b6"}
{"id": "train_036635", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "4d739aa562ce44b97ff84fa35893c990f79397dd"}
{"id": "train_036636", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ae28d2c2c78be5a3ed03257537071118d881771b"}
{"id": "train_036637", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "3ffa74404982d62c4d68e6a83ff987cfea7e4181"}
{"id": "train_036638", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "354083d148cc9dd93976843f62ef5d65f586acd2"}
{"id": "train_036639", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "36d856e33320ed97ab6a959032ee5552f25061c3"}
{"id": "train_036640", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "355b8a96f8191d1159e00f651f54c97f3996d32a"}
{"id": "train_036641", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "0247b493cdaf649323b959428413189fc9f36f2f"}
{"id": "train_036642", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f1705c3afe9fce8538e268a86948df5956892cd2"}
{"id": "train_036643", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "554e13908edf3f62971a334a42eed881dfd48747"}
{"id": "train_036644", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "53f7ecb79abcf4dc819c461a06c8a35b7404c564"}
{"id": "train_036645", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "d039862e6c62102dd68b679dd08c115862ef31d1"}
{"id": "train_036646", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "1f302e82f031ffdb9c14bc7a0c53d5cf83de61c4"}
{"id": "train_036647", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "7875d759c8eb131cf8f9efec42252ceb0c70cf7e"}
{"id": "train_036648", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "5a886033f5c472fa095425e06bd072e5c5a2e031"}
{"id": "train_036649", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b92a89ea53a982057e7669c3e76ad1d9ee7feb94"}
{"id": "train_036650", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "b386d3893531e2f9d1c36bb3b7a008ab9d88b33e"}
{"id": "train_036651", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "d45bb2d5ba4dd3eec842589cdfe0cf7fa9dc7747"}
{"id": "train_036652", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "cf3f565fb1bb923583eaab5cdd11aafdacf746ff"}
{"id": "train_036653", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "af9fb3c897591526de2fba8eedfd64dcf75479e4"}
{"id": "train_036654", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "f1ae743d374f8df4f2efb517701d8adec67248ed"}
{"id": "train_036655", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "f230463b196dfc7a29234a13cf4d0c2fd742b75a"}
{"id": "train_036656", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "4bf336d50b5b63bd245f47a4626595f3f63a1ecc"}
{"id": "train_036657", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a05fcedcc8f4f4cba36dd37d4a1919aa92e34ff4"}
{"id": "train_036658", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "299282ae16ae2dffaa87c4bc34b4080d967fb3c4"}
{"id": "train_036659", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d03118fa8aad4477dcc6cc06e31e6a1af68e580b"}
{"id": "train_036660", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "67301215ccecd0b60cf19135ad9b37dc801a1091"}
{"id": "train_036661", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "5ad8940ff2094e8e766ec54f15cf2077db5c926f"}
{"id": "train_036662", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "b0fa9aa3e5685b7d23cf08ee5c054ced47581b02"}
{"id": "train_036663", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "bccce48425457f7b9088a66ff11090680e749316"}
{"id": "train_036664", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "56a1ff83897c9e326a7ef5f8d52f0017f50ad18f"}
{"id": "train_036665", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "5d1e0453c7c5e8a86e47e9d56c0d4ef132609dcb"}
{"id": "train_036666", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3489d8f49a796a69d8a222a94a9a486dae6b794c"}
{"id": "train_036667", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "90e0b8d0321d4c7d5be6fa67f17473f5e6642d59"}
{"id": "train_036668", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f91d3e213a17c8314a3a71463b3fcacb9f8e45d7"}
{"id": "train_036669", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ebf9703928c7c6ed76da7f7beb3a982246825c3e"}
{"id": "train_036670", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e2b67df2b04b911826517f0a6b6f1e2049284516"}
{"id": "train_036671", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ea335da809db4488cf67dd5f93f5ea4bb66def04"}
{"id": "train_036672", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "bb5b38717670d43ab9d781b0a930afc639748eea"}
{"id": "train_036673", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "ff6294c040e7b04a1d62f61fd35bc626bba4f892"}
{"id": "train_036674", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "79e9dbc8b60db978210af421f6f7dce5d0384dd9"}
{"id": "train_036675", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f8ee6f41bed09125ef44547fb099b05fb610a219"}
{"id": "train_036676", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "7af3bb1a4025796c2bafb6033bf3ce75c88f42f8"}
{"id": "train_036677", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "ecbc9e0192f8d3151bc9042d6a0082f71e3ba0ee"}
{"id": "train_036678", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "3a073500ff6b8a1efa524f51c57c51b743ac7544"}
{"id": "train_036679", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b4312190f435ea9caf2fd4ed4c42c4c87e0a31a6"}
{"id": "train_036680", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "7e16e57757cbb0a6bebb49d4ffe2e2abedd9d684"}
{"id": "train_036681", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c5d75035739fc009f291f4ecfa77984185243783"}
{"id": "train_036682", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "64639bd9d60c85378330418071d5291722392384"}
{"id": "train_036683", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "1f970eb1d1ed6221446cfdc57377ad737b1069e2"}
{"id": "train_036684", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "2d3c0f3a12eea90b0d3317010a9cd7b9c1eaca76"}
{"id": "train_036685", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "3cf43a623aae40a59d58f9329bbda9180b07cf7f"}
{"id": "train_036686", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "29098a2e3be54b6dda1753b52570d594d1be6757"}
{"id": "train_036687", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "916890bcb5e40f6ea5d402940ffb9f7fffe13b68"}
{"id": "train_036688", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "593b7fe67f97102c02694a8d1c7d92a25ae00602"}
{"id": "train_036689", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "60dfb181e06eca47679a5c0f455dc9bec90fa25a"}
{"id": "train_036690", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "3af8438c7466ff5c9bd4d7b595ee5ca6400e6d23"}
{"id": "train_036691", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4a26b4f7c4ad2d8a00f1ae3aa2d0f0e21d6044d9"}
{"id": "train_036692", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "93691cbf3cf750f24629f60fffec29fbff5d50e0"}
{"id": "train_036693", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4a6430cd8f6821eece3b04e6fa4fdb95bc692c60"}
{"id": "train_036694", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "5232abfb9176a2c13741df0c3efc301fafbb84fc"}
{"id": "train_036695", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ab47c9706933abdc436780ecf5bbbb8a259c895c"}
{"id": "train_036696", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e7db194e4dc84f4d2f6b85b229327ca2dbd00eb4"}
{"id": "train_036697", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c973a2cd0391419e3c55535e1c8e70cafea0ea0b"}
{"id": "train_036698", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "9c8675dc0448ec7bb0f6f491dd9c6cefef5d76f2"}
{"id": "train_036699", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "069822c6603271a45bb76e23642d671817166eb7"}
{"id": "train_036700", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f4e290e1cdce44c4bff8efcf81b0851fd2055a80"}
{"id": "train_036701", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "10474ce76972461957f2559f3315ef88cf402ee9"}
{"id": "train_036702", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "9eca8a806548a838b75133a4898a28014718c96a"}
{"id": "train_036703", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f0cf42fb4d85485f3d858eb345f0a04617e6641d"}
{"id": "train_036704", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f18410317464e1b58b7d50627cb98f8e52186190"}
{"id": "train_036705", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7a4acb9b0a00bb369c16befd42622914a6cebd4c"}
{"id": "train_036706", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "44383ba935ee3a956a3970445d190b2e6609d831"}
{"id": "train_036707", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "0b5f078931cb0220c1d45cefb0a671ef73aa0e1f"}
{"id": "train_036708", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "fac82d338b8dd5b6c6c99dfb1ac265c2d7a80974"}
{"id": "train_036709", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "6166d139a279bd719cb3da3694b0d565d3d672d4"}
{"id": "train_036710", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "fb63e8f9b9d43ed76eebfe17268f5c2f92d1ef5c"}
{"id": "train_036711", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "eb33ade524f3b39e81364cef2fa07b037f87eaef"}
{"id": "train_036712", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "8dd55885c2e6ad1389c7a8872807e0819b23398d"}
{"id": "train_036713", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "8f0c1da27bbb58b85c480276a97682074363373e"}
{"id": "train_036714", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b9e9f3be9b77d8c8f75cc2e695a1443fc759669c"}
{"id": "train_036715", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "2e932b57a5369934d2c93f012fae15fc5dfcb4a0"}
{"id": "train_036716", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "8a180488f06b7585aafb526c3691da2d7a24f7b5"}
{"id": "train_036717", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "28881d5e67c93960837b7529affbf70057378c52"}
{"id": "train_036718", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ed1e491c4172f7fbe9ffae9ad43972ecd0d33ca9"}
{"id": "train_036719", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "1540d558ffcd7a1b6df058864ec75a7c354f7d14"}
{"id": "train_036720", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "86c66d35a2115ae3410eb134a7e58e4b0cd4d654"}
{"id": "train_036721", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "7ce203ae0d1b7558380bd5f327cd4155265638af"}
{"id": "train_036722", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ddc0274a1f8360c842f9d339a04fbb2d50444b9c"}
{"id": "train_036723", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d3baea651d689b1512b7b8d79ea028f1de5c3834"}
{"id": "train_036724", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "6692b5a5fdcef32aeed96e0fc4723fbbd3e4c00a"}
{"id": "train_036725", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "49f4216ba43021ecff8d894620f56043b8690d67"}
{"id": "train_036726", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "015c91103adf5edaf11e56d620c344e9275d029f"}
{"id": "train_036727", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f715e45cb29c631e73471e27187d4e9a011a43fc"}
{"id": "train_036728", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "d0fc52cf220dc6b7b996c7e0a22c9f7e038aa9db"}
{"id": "train_036729", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "8dd26a9928e5d3836faa5c0a15cbfed0b82dd784"}
{"id": "train_036730", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "dfab1422c1e0dd36b75c344dcd4190685fc8427e"}
{"id": "train_036731", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "34c88047ce976cea28b9f2eed2841e148edda472"}
{"id": "train_036732", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "8c4c8e7000d98e38ae32883ab5b24c4fbc89f689"}
{"id": "train_036733", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "2e6b23202b348213482f9de4866033fd0513d103"}
{"id": "train_036734", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "cbf836ec91ce3fa115604931373a516029f606ad"}
{"id": "train_036735", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "44a101213b078a085426d268a9a7bcd35e6af5a1"}
{"id": "train_036736", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "236533a40e2c9002301bc23cf36bd9193cb0aa64"}
{"id": "train_036737", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "8ad98081bf157e34a7696cc2234332ca4d67073b"}
{"id": "train_036738", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "2b202bdaad9ccada990689b9d28ae6ae60ce67aa"}
{"id": "train_036739", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "89524bc5fd1d084e19660453ae870c49cb8ef09a"}
{"id": "train_036740", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "505762040219f1659b725a342e8ce863879da074"}
{"id": "train_036741", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "12b5ec1c6aa2b7dc9d2c0db68cf8f0ed5ed2d689"}
{"id": "train_036742", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "17f0f2b262db31f74293fccb5799854ef51b6974"}
{"id": "train_036743", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "e643a405f526fffeebdd8f3673da0c6031a3deec"}
{"id": "train_036744", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4c960e2b387c35c26eff82016a33f796680c2b85"}
{"id": "train_036745", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "3171d39d7afbf66e23fbcf4b8bc2bc2cb6dba98b"}
{"id": "train_036746", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a5bdf87aef9b7abc3fac2ad12187d55c6096817f"}
{"id": "train_036747", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "8e7b13782149e8534cf5edf8eed541080630bb87"}
{"id": "train_036748", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e0e2b9efe4dfe32223cf5667692a640b841a7257"}
{"id": "train_036749", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ee933905944756175711744a5552640621c7d002"}
{"id": "train_036750", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "0343e18203f78148940e12c148ca9f3d4750b100"}
{"id": "train_036751", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d0fac72958b9437ad3c87f8cf75a4e66044c4eea"}
{"id": "train_036752", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "4f783c79e8213c6b0d35db53eaf6e18d8e4f4a15"}
{"id": "train_036753", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "5847ac4d08a03e24fc826788d98c81c66d2b8dd1"}
{"id": "train_036754", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "366b5b568b3a894744eeab363593fb047c7bc24b"}
{"id": "train_036755", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "4260092749a7eede393c6b9fe841b73289d962bb"}
{"id": "train_036756", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "d1431c5d5d9bf218aee134dbe777af36796277b3"}
{"id": "train_036757", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "e57ca0689b8494c2120d12e529c07a2595e541ef"}
{"id": "train_036758", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "e38ddaa8a8cadeb54668087a9aea9b84c1b8b0b4"}
{"id": "train_036759", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "166dee65d20d951dc516501d45b649880fab9637"}
{"id": "train_036760", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a25f92c009eab7dcb1369445ea035a50b2836a35"}
{"id": "train_036761", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0146d43e27a02c79f6669d881a73462fe84072f3"}
{"id": "train_036762", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "9de7345bb3d5f4b154eec762a0f4f9bc6b4a3d2b"}
{"id": "train_036763", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "8a2fbde8cb6d3e4a13d2a06c37c8d59509dde24d"}
{"id": "train_036764", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6908066078580ab42b46f19816605708e3973eb0"}
{"id": "train_036765", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "f0d45a0d7c221c0dc2ae281b11b7416a4430db79"}
{"id": "train_036766", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "ec6e5f8d760db1d14ca9c8ac441feba98bc963e0"}
{"id": "train_036767", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "3aa6475f611ec4f8628d598d6789f8986546559a"}
{"id": "train_036768", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "69c624b1e45658cb04eb8801d613386d72be7683"}
{"id": "train_036769", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ce5d647135b99d86136a9c57c3c9aa1510af2052"}
{"id": "train_036770", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "9f337afc8a6a63e96fb605bf59ea6d90f13bd50b"}
{"id": "train_036771", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "f091609845e002d2a49ea2eff5f0ca7460160d5c"}
{"id": "train_036772", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e512cb7d3bbca18c722a2e5435d06f2cd5a583ac"}
{"id": "train_036773", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ed4cf7da4780ba1390b461af2a7a00ef72646d19"}
{"id": "train_036774", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "21022fb3c1da2f03551dd9b21a70787e4b053593"}
{"id": "train_036775", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f1d8a04b4541d9f5760a26306e4c3dd2627a1b06"}
{"id": "train_036776", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a8e9f5ee8f57e09b6cd3b3135d6e0c0d41b895fd"}
{"id": "train_036777", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "89f6ec044b894fac7663f99a2738d3ea0ca1211f"}
{"id": "train_036778", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "21b67d90a61b843c1fab70cd2ea0b43a9b6e9d8e"}
{"id": "train_036779", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "5d11d822c01862345a6709a94a20b25ff3693375"}
{"id": "train_036780", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "fe886960abfec00f4dd64774d73a378787731da5"}
{"id": "train_036781", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5c585d2bf53bb4072b0da563f3bb23d57022d291"}
{"id": "train_036782", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "ed761de5f826c6bd1a215bd654f1bd6b54f0fbc7"}
{"id": "train_036783", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4386d72f48fe50b72d10e5af82ec663f54bb21cc"}
{"id": "train_036784", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a2ce7c08882776fd71aeeab96449e271b4ea9139"}
{"id": "train_036785", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "b3dd0ad993c51363b9578a808e1f6e037e0e50f4"}
{"id": "train_036786", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "615f4dde4ae7f5301d645ced13161ab4ae542514"}
{"id": "train_036787", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7e44f0cc7f963dda0491395d321f5a2ed14ddef9"}
{"id": "train_036788", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "5eac2a794394ccf6deeb816a9f8bcc7eb5ba0e47"}
{"id": "train_036789", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ef670f53ea84c22c5811800271dcfc8b67ae0870"}
{"id": "train_036790", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "c19e09adbdadbaeec7d90b4e3ef9d595a8a84fac"}
{"id": "train_036791", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "36dee19a2520052f5e361105d38159379bf67df9"}
{"id": "train_036792", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["agent", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "0edf02213cb2f8e6382ef7ca720ee722a094c968"}
{"id": "train_036793", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["agent", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "73cc2ebc495b1a5388a3eebf5155285b6a19e1b8"}
{"id": "train_036794", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "8b49aa6c4ef8563d7c53be3e04e336227d4d7ca3"}
{"id": "train_036795", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "3f1c5e5695b3aeb50ee795fd124960e6272e7946"}
{"id": "train_036796", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "144379d37767c9b0c8b609782e359489f634b7ce"}
{"id": "train_036797", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "ba0fab82e52b5366546ae89ce2a6de8eea1d86ad"}
{"id": "train_036798", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "eae673d632652a0eafa9d33e197fa83f17a1e981"}
{"id": "train_036799", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "9e4b02096c13f4b5eaa71e2da4ce7ee9c25c3d97"}
{"id": "train_036800", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "d04b0cb2ef777ce2a2c44ba80c4b725827ed6254"}
{"id": "train_036801", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "871fd0d1f2343a0a6005548d49fcc0d4e16b5662"}
{"id": "train_036802", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["agent", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "a39e80c91db68f9fe6cf87f2acb9d44e8d807906"}
{"id": "train_036803", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "439dff1a2aad2180bbc67dbb7b42d63192fd9685"}
{"id": "train_036804", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5bb3aa0b193eb1a383f6166b6464304456d5dd7e"}
{"id": "train_036805", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a246dfdf203eb52462f941376be434e3f40584f4"}
{"id": "train_036806", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "64a655ffbf298d843456b63f7bf1c3216d057de9"}
{"id": "train_036807", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "57e1a2f18f725d2cba3d59492884cfdc6ec7920f"}
{"id": "train_036808", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f91d3e213a17c8314a3a71463b3fcacb9f8e45d7"}
{"id": "train_036809", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "8e2b17f448d2763a048d5893b564487496c1c96e"}
{"id": "train_036810", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "87e9cca14f20c21d372e8f2a9024733735d82525"}
{"id": "train_036811", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "cdca2f1b9c93b8c81022fdcc4dca4fac44f09388"}
{"id": "train_036812", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f05c3ce1b913713ec7dab3eb33efd872f971595e"}
{"id": "train_036813", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "9ac78873b5030e71c0cd58585c9b2a57fd423e90"}
{"id": "train_036814", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "cad4ea5bfe863221fb55d2a13367f683b8c83eca"}
{"id": "train_036815", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "748a6e819da525a75a09e24322ae425f1c8703cf"}
{"id": "train_036816", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d56920b4b6b22cbc4aba993d249e883f242ebb61"}
{"id": "train_036817", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "6524bc73a2760ab34a274cee9ac4d4d6a7ed54f0"}
{"id": "train_036818", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "47dbfcd8f185dcf74d55cd5af473b5d442b63a67"}
{"id": "train_036819", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b93812e2c50be8c3605cb16542358acbef6e7394"}
{"id": "train_036820", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "705b8a594ce5de5c845c4f48912abedb4a7cad62"}
{"id": "train_036821", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "e91d3aaf89dfc3a576f105bc7010e3b1a9d2e506"}
{"id": "train_036822", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "25e610d39e4d2070f216b75f69fe87d9d00347fd"}
{"id": "train_036823", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "2b4fd7c5b917e28ea9e4910d0e44eddf37b4e692"}
{"id": "train_036824", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "bf0413ac73e73085dd8f272b10e986a8b92a7807"}
{"id": "train_036825", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "4bdfef78a48acbed494bcb74ded716bdf7bdd9cb"}
{"id": "train_036826", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "4bb200a8f7d27b39a74dec9850416072c07bd302"}
{"id": "train_036827", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "af6a99e42332b625f573a8801853d3e3794b9869"}
{"id": "train_036828", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0a7433e4b5789ebdcd9a684538706890fe639816"}
{"id": "train_036829", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d9f16676c72163508038bf2c2f57cafba59fcd67"}
{"id": "train_036830", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "06cfe52ee7710bb4a7ae086a4c3b4600206a4850"}
{"id": "train_036831", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "6fdb4a24ab4d476dc97bc913a155d0c5eb9e2127"}
{"id": "train_036832", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "00a039d8f2879e5014d976966c4d424eb1bf02ea"}
{"id": "train_036833", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6403cc058490676cc72f0c3cbe63d768d74f4c71"}
{"id": "train_036834", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ae26d08fd1cc3bab6d5a24ee922df708038cec0f"}
{"id": "train_036835", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "5ff8a55f5217e205ba33e2d57f77306407c8d0b4"}
{"id": "train_036836", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3ffa74404982d62c4d68e6a83ff987cfea7e4181"}
{"id": "train_036837", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "af0536687d659b46bfe66aafeaa9644d974de42a"}
{"id": "train_036838", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "128b26e7038fecdead43aca7cc30d66ba9dbd9fa"}
{"id": "train_036839", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "86db11b10ed2dc4a5710c776381bd19d6fe4516e"}
{"id": "train_036840", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "122e0cc1d8cd6fad6d34b75d5af713eeb5bdb45c"}
{"id": "train_036841", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7b8df5e6caab5e1a62fd2c901abf3b336134bd9a"}
{"id": "train_036842", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "d128a99853eb78d08eb6621ac187811480dd7dd0"}
{"id": "train_036843", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "4898cf3e38bcec89b1bb91297c92dee10dc46651"}
{"id": "train_036844", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "cb240f13c257c7f8eeae0f7057a4a2fa9912ac33"}
{"id": "train_036845", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7063e9ffe5a8edd9535a11e45290413fe9fdfc41"}
{"id": "train_036846", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "43221d1be9393b99a0b4922690517be290a3d6be"}
{"id": "train_036847", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "f8c335719622279152a56d563cf0a3efbdc272a1"}
{"id": "train_036848", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "29155a1d6feb6bf10cff1f501d693b19c41ea2e1"}
{"id": "train_036849", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a4a91a5e35c5df47a4074c31b2af531068ad9e62"}
{"id": "train_036850", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "160463cee8d58546990e1076072b806b3d14e76b"}
{"id": "train_036851", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "56a90080c1cdd99a93fd6ee402156a23db4fc947"}
{"id": "train_036852", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "21879f04c06e86ca02e0e745170c7b86c4b555bf"}
{"id": "train_036853", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "cc6901d324f723e4fe22d0b19c451aad0d8dda2b"}
{"id": "train_036854", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "0dc36facd33d81e097712a3b6296bb981f1f061c"}
{"id": "train_036855", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "c3575b4d6b302f14cdfff28f4987e99c9bb564e7"}
{"id": "train_036856", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "1430fb6b7ce7e2cb070f66ba05329cd6febdca7d"}
{"id": "train_036857", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "abcc72c89e0ca12e99ac373645463c5eddf6379e"}
{"id": "train_036858", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "b7347567a406b6add32c5c238dee632477a69814"}
{"id": "train_036859", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "bb0d9cedc3693ac4d974cab49a123c0b583517d1"}
{"id": "train_036860", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "405a5d4865d00bb1c39af150846de162a17f5e9c"}
{"id": "train_036861", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "3f452d943186318c1d7cb908b538818834e84037"}
{"id": "train_036862", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "8292b61a2cc9e7b7c85f82b7f7647b3c4fc76ec4"}
{"id": "train_036863", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a6e6bb42214aba0e44edfd532c420b686f2ba908"}
{"id": "train_036864", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "d76831f49ac4404a16ec22aa901067916059ac75"}
{"id": "train_036865", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b11686bcd95217d5838bfe01912f44d67458f177"}
{"id": "train_036866", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "648a85ab7b8a6a189e3e684fe1eaf4957dfde675"}
{"id": "train_036867", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e5360dd6374959a26451f15c57ffd0c920becc28"}
{"id": "train_036868", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "1500640766415c5c43ff74dbfd245cf4602dfd19"}
{"id": "train_036869", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "df8991d850f9093c9b00323c892d02c050d33673"}
{"id": "train_036870", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f11c4627fbf44e0d9efef93ec961476cc1cbda13"}
{"id": "train_036871", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "ad5070e71080fec961d632d2be971289c10c9f3f"}
{"id": "train_036872", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ad6a135469cd0b0cde64e74aa8c39c414e232245"}
{"id": "train_036873", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "cc24267b8ea39ee7b54891b764cd13c9fcf57480"}
{"id": "train_036874", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "0c4b164cf4996d49a7ea6b1d2fbde6e20ba18c4c"}
{"id": "train_036875", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "15886ea4b8fbce078762bde40bd2145ef1100a53"}
{"id": "train_036876", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "7264d42e56becd04ef0d270a2ef8754720ea8ca8"}
{"id": "train_036877", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "3bff15f03f20202bdb26efdc979cbbe51bb1b3e7"}
{"id": "train_036878", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "694d959dabc9905fc76a5da3fdb8c7535d020911"}
{"id": "train_036879", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "7437bdd16d801446c598fcf3d5e033724c253476"}
{"id": "train_036880", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "d45db4bc6395e3668f4ec585636d829fc184112d"}
{"id": "train_036881", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a5ef3b77a1986aed4d01ec72c9d8142bdc2af400"}
{"id": "train_036882", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "64fde3e3334e694b5106483b54aa87e467fa88d7"}
{"id": "train_036883", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "be9e134c9ac31073aced3a7b7be99f3870c40401"}
{"id": "train_036884", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "38433fbdc5271e880f61f76f86c8a0ae0586e01a"}
{"id": "train_036885", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5b505e2b78cca8fa4bced8c07fc87f8e78c5260e"}
{"id": "train_036886", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6dc8a7933fb793ecb6e8720a48780ac683bd4c5a"}
{"id": "train_036887", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "8426b927961f98bdf9fd64c85e2197550d6d04be"}
{"id": "train_036888", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "002fb5656dc3a2d3425099ad3ff79247ebfd9bbb"}
{"id": "train_036889", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "0ca87ae4272f6a4f32d5db207cc8f943f4b1ca2d"}
{"id": "train_036890", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "627976cc6be57eeaf309f9de2d9c3577f03e0229"}
{"id": "train_036891", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "4ee57f8185199a3b090f26e3a53a34056d6af6f2"}
{"id": "train_036892", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "69b04c417f2917376431d8d54f4dec28ee727415"}
{"id": "train_036893", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c40a5689bd2eb7e3402bf13878698ab2fe73dccc"}
{"id": "train_036894", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "6809421238dc843a5c237fa3a673f219e111506d"}
{"id": "train_036895", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "8881b11d6dccd4310274e828ee7af9bc53696952"}
{"id": "train_036896", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "90cb27146f11f4cf4c0639916e935fa33e94ea93"}
{"id": "train_036897", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "e482bb2f836558a80911d13551b6b8306c5bf523"}
{"id": "train_036898", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "9547e6ccd3b1bbb049570a0c265cc6a1c657b27d"}
{"id": "train_036899", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "fa8c065375637fb28fb55f68e326b04f3c495460"}
{"id": "train_036900", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "271ac7f204d8b9804ab9ef1e1e0e040b9009367d"}
{"id": "train_036901", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a6a0d0eaa3b8eab9efb7e732abb5e12ab933c50d"}
{"id": "train_036902", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "550c4d7176fbb91170501aa483075278b3422ab9"}
{"id": "train_036903", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "2ed89df55bc933a07e7b8cafde0a71a6bfd8f67e"}
{"id": "train_036904", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "0ce97936c9601d294cde82e2b405407efce5c112"}
{"id": "train_036905", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "989627840ac593f619cee67af6f5b218dfa039af"}
{"id": "train_036906", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b93433af064315b64b1107126888e16a5d186b71"}
{"id": "train_036907", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "da7d65323ef2711cc1f031fd46ababe39cf20b57"}
{"id": "train_036908", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "2c366f71f837efc9c05a8555c78eaea59672d69a"}
{"id": "train_036909", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "e4754167016a7a00431316b590c802ded08d1790"}
{"id": "train_036910", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "53f0910c2f1c2359240f3fe1e3d1786c8c5a1656"}
{"id": "train_036911", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "fda254ffec9499aa8b8f0e3c47d14a3c3b8ed085"}
{"id": "train_036912", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "21871394db3eb311c8a4663788c5294f236a6eaf"}
{"id": "train_036913", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f597b21f4124c3354007c102eb4b45d37c65057d"}
{"id": "train_036914", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "529c2da6924aac04a914021bc1ab36677bcc4660"}
{"id": "train_036915", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "ca3a2760eec83643396909c0955c1ed30dd7a017"}
{"id": "train_036916", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "fd9bf215555533014091a54040e3a9ceec53aad4"}
{"id": "train_036917", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "efef004a05dddf7b2eb1232c193e9af53cb953ef"}
{"id": "train_036918", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "cf6feec8b15a87c5b56022e4944c49b9bc02f25a"}
{"id": "train_036919", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "700be01d769964fc459e878d61f204443f36ced9"}
{"id": "train_036920", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "fe7927178d428fb3ae33dccbab48b1b6fab582ba"}
{"id": "train_036921", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "e2cf6319b6f34e152774c9a5221d8df2108bc638"}
{"id": "train_036922", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "7b0fc61f8ec5305f0e62dedff4fb7a170af3c2b7"}
{"id": "train_036923", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "972d6a461fd2b8da360f3fd8629b929392038600"}
{"id": "train_036924", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "388ce54f570020eaeb85fcae4b8b7eec041af4d6"}
{"id": "train_036925", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "c8db01edbf25496ad99a842357e7e56c46ebc5d3"}
{"id": "train_036926", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "4b209c59cef8a3fcfc5097c9ba887d88fdc95073"}
{"id": "train_036927", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "0d7d9386b2175b90174ecfd435b0f5492d0cea1e"}
{"id": "train_036928", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "c1fe18ad108caf540cab3f17383d1d7ab09b9094"}
{"id": "train_036929", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7f1b0a0340df55c00d4ed8e9fbe79ee10f395ef8"}
{"id": "train_036930", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "d59cb81588525db967ef28ae084c770adf82f9b1"}
{"id": "train_036931", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "e438b48d180750de3c5ff71f7b1b0143790fb563"}
{"id": "train_036932", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "757e6bd621b77b55e1ae9b683def78fb8b904514"}
{"id": "train_036933", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "3a1bd14dfae9fad2e99b79a1b450495bf707621d"}
{"id": "train_036934", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "95244c96a062804d063979afb89bb03245ba004f"}
{"id": "train_036935", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "63d448acaebe1f7a8a406914cce79fe541216d79"}
{"id": "train_036936", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "eaeca5c1d241349cec1fdafc0964bc577e35ba06"}
{"id": "train_036937", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "757d938eb113915c294cfa97594d02747b0b5643"}
{"id": "train_036938", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "87c52ca7d4d6ed391134acade9600112ae57fc74"}
{"id": "train_036939", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f8f9a5cd3e54ca4dbbf28d3f82dd61ddcdfb3fc2"}
{"id": "train_036940", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "9aaa70e06f47738965233c2c63d63d3807a39df9"}
{"id": "train_036941", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3a4731de0a9cd47b79151c79f1a64443a3428c72"}
{"id": "train_036942", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "1a883e1cbd77479c6ddd74463d9fc5c23a95aeec"}
{"id": "train_036943", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d4152a9e142d577af002fc9fb87bb2ecf3506134"}
{"id": "train_036944", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "28a839eb6f8599be8bba44f1769413025a46f5ee"}
{"id": "train_036945", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "ae775c0ae7695aca53f63a8aa3404e1cd2cefc37"}
{"id": "train_036946", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e6a9f080c9d78fa1508fab4fc454958b3a59439a"}
{"id": "train_036947", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ca0c8475da60b1db08044075976dd9b38228ce0a"}
{"id": "train_036948", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "9f7b300e83e75f76b591bb5535c1f44fa33a35f3"}
{"id": "train_036949", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "a4deda3c0e690f67d5ad837b13412fb48f438bcf"}
{"id": "train_036950", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "2f7c70925f3a5221e5a823edbf3af3c6b481f368"}
{"id": "train_036951", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "60aafa89d07309824a74c30f18e365e4a1bfdec0"}
{"id": "train_036952", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8210f02dc9813ccab5e225d4e99811e735801be2"}
{"id": "train_036953", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "1c432db2422e8ed6324b6f99e2204e49719c6776"}
{"id": "train_036954", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0e096a8c8df7991b6325f9e84e25c037f5a274d1"}
{"id": "train_036955", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7c79a0ee144dddc42076f41648575e0bd84c360f"}
{"id": "train_036956", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "d08d148cde4f2bc6a402ab06537cceb59c98f0dc"}
{"id": "train_036957", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "5c34e0aa405d194054c29b15650a20698cb87e98"}
{"id": "train_036958", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "413270efac9c51829c7fe1217d5fe61bf1083644"}
{"id": "train_036959", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "bde16e8efefa1988e3b1c83546d1a17d80f74aa9"}
{"id": "train_036960", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0511033023441f519ba6545d78de82e5fd688b20"}
{"id": "train_036961", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6abd4c26981ff504675a568820fa73b51019c707"}
{"id": "train_036962", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a3ef192a82f1f87fc4d113d8350623eaf4e8490d"}
{"id": "train_036963", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "66322ad4f646a58fde2a86bceb04bedaaca2e223"}
{"id": "train_036964", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "a4539f69983dcb084a10c5367b5d9fd90d65ee66"}
{"id": "train_036965", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "82554f63b9976e48f5728c985d645325306d5eb9"}
{"id": "train_036966", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["performance", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "d5afc75841e181f166f09fa6e0dd2333183344a0"}
{"id": "train_036967", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "8d360de7f18d067ea53feb326a5a62a4988d5fc9"}
{"id": "train_036968", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "065236348e0e7db29b7aba300edaf0a375321130"}
{"id": "train_036969", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "1b7bfcb7f8cb976f6fca951a9559e2f94b37d56e"}
{"id": "train_036970", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "35be5061a305be125e8772fd636bceda7d62d76e"}
{"id": "train_036971", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "42a49b1a080770587de92d57312d896996401a5d"}
{"id": "train_036972", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6875f4770c387a66d241139663407ab8168c0566"}
{"id": "train_036973", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "e6850e89a2cfdca98a009d2205bf583ceb6140ab"}
{"id": "train_036974", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "698804ee6110287cfdff4bee04787388b80ad193"}
{"id": "train_036975", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "24c2ccd5cfe97460623e91e63339f34e1575ecfd"}
{"id": "train_036976", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "1add01765b0ba2c61c7c2e782b967fafcebe458d"}
{"id": "train_036977", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "4b75d31c3d22775b6e33d685e953ebc63bc1fd9b"}
{"id": "train_036978", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "e255e81c644bd5850aacab07d61ef44758e6ca2d"}
{"id": "train_036979", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "803972cd103600731eea9f556348e9ae809c1d03"}
{"id": "train_036980", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "50a3b1725bf5543907208a5fe63c90dcb1823178"}
{"id": "train_036981", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "90dee186f8326d8b0cfc673401d84acb59fea014"}
{"id": "train_036982", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "e9a26e37bbc6cf5beb7e0d7f81264f32e5361e46"}
{"id": "train_036983", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8e704297a26319f0c785e4fa32443c04b2eb4c25"}
{"id": "train_036984", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "efa676a469a878522548fb268817cfb0eb03c6eb"}
{"id": "train_036985", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "681eb3fae676401c84592460bbe48034b84f6291"}
{"id": "train_036986", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "7aee87d911789afe1cdbcdc043539d165ef2260d"}
{"id": "train_036987", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "0ca8aadd21f41835fd1a6f8125f110f64b945ba8"}
{"id": "train_036988", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5e03356b68cbb94eb2c5c9d6bb87ea4fe081e802"}
{"id": "train_036989", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "7f0806dd59fd55133743d227aa0b229211927e01"}
{"id": "train_036990", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "8dac6d481ef7ca8bc13532c01f900b30fae1e786"}
{"id": "train_036991", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "0924f777eebd3033d6ae531e53ba9012e4a683f5"}
{"id": "train_036992", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "56df551d69997ef9d49cdf7156c65b6f25b4488c"}
{"id": "train_036993", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "321072ec108a0475034a422862d90f9afd4d5559"}
{"id": "train_036994", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "75dc639526f675ab1b257fed87fb5f4e4ed867f2"}
{"id": "train_036995", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a7fb6fb23e639d23cdfe5599b3ba659560901da6"}
{"id": "train_036996", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "7cab9722958ccd0f4d6c8b1e126ce57ece0471bf"}
{"id": "train_036997", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "5643b607e2566d4c01894e6779fe6817e40ffb65"}
{"id": "train_036998", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "4de414589c88046f4e9d8a923825021e2fcc50a7"}
{"id": "train_036999", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "89d5bb1bc0debd238487c31d2f6362e7846c35c5"}
{"id": "train_037000", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7d0e88653621029c26c87ac776dbba54a0e081a1"}
{"id": "train_037001", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "61394271a259bc80384d0355e0c97c0f8ffe7230"}
{"id": "train_037002", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "36dee19a2520052f5e361105d38159379bf67df9"}
{"id": "train_037003", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "f544c340623b25bde89f8ffa605c50e6411fbfa5"}
{"id": "train_037004", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d33068114ba8b1d9e19781d5f98a6e477e50c025"}
{"id": "train_037005", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "700be01d769964fc459e878d61f204443f36ced9"}
{"id": "train_037006", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "7d332eef9d12ff68a5fd61b69e0f228c50588759"}
{"id": "train_037007", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "515b368fb22e51a2512d2cfc27547d6a15ad5ce1"}
{"id": "train_037008", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "eeb1891fd74cf644d052875aee999eb68024fa60"}
{"id": "train_037009", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "df52cfb54d55529430fb514ba61d5c815271b746"}
{"id": "train_037010", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "847e4030a0d3b7a8cef1bf2034d71ef04671bd28"}
{"id": "train_037011", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "3140fb506beeb63fda7a1e7acb2e59ed193ab087"}
{"id": "train_037012", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "58fbde1fc0495fc3821077f04594741cdda4487f"}
{"id": "train_037013", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "65f41958e0e8b016bfc1d914ea4612b90b662270"}
{"id": "train_037014", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d9d1494b6ae424641a7faa86950bff8c7d9907d9"}
{"id": "train_037015", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "c1546d37d804673c3530bd4527df9d327fdc24e2"}
{"id": "train_037016", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "58cfe03903c5b46297bf32acdd1ddc5a3451db9f"}
{"id": "train_037017", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "7bde3290ab656dd7dbace389ca1bc167699060c7"}
{"id": "train_037018", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "8b26ff2da70de2af107a91f66cfa25e84228d049"}
{"id": "train_037019", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "bfebad7db4d8db4d2d667dd02d27f657c7a57a79"}
{"id": "train_037020", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "417ddb70a9166954ecfe95c87a8df123586cdb23"}
{"id": "train_037021", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "8cf920a1f6ee32e8e70df34de68d2bcfd9e5e321"}
{"id": "train_037022", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "cbeb7691dd0726feb4bf549ebde21a1b395dfb03"}
{"id": "train_037023", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "f1a44e12718a814fd8118b9bf1ef356b17aebc45"}
{"id": "train_037024", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "03d493f27febdc912abf2442ff337d76c658879d"}
{"id": "train_037025", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f4df679c7ba202de10ae072a05117b8d2f57cb8c"}
{"id": "train_037026", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "b1443d496473e621b0be8bf49d7e8403a9a25041"}
{"id": "train_037027", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "5bb96184c143f6b0addd4045ab98cde1620079b5"}
{"id": "train_037028", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "6502d79da12be17c773583b437f8daf424e3aa55"}
{"id": "train_037029", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0a7bcf4c0d35f2e1e902fbf5479b4017e3853250"}
{"id": "train_037030", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "fd0e2e02b23ad66ae419950c2b1bfc70ae64d6fa"}
{"id": "train_037031", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "9435b31d62a85d61db88c7ffbd273e24ccd68f41"}
{"id": "train_037032", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "be198fbbd5051a98696a46e1e9879dfa096dacdc"}
{"id": "train_037033", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "e7cae698e2bedf062a59dcd2b0654158adf2c19a"}
{"id": "train_037034", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "24fe50de8a333abe903220ffb16cd860778e586b"}
{"id": "train_037035", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "0a7aa09f9c75f45732f57808c5fce66c79191e7d"}
{"id": "train_037036", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "379f7701cefb587053d317f92f7732b098d024fd"}
{"id": "train_037037", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "d1e6fa577b1d18d2b9c40a74178635b3f1446a01"}
{"id": "train_037038", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "94b6b014cd6cd93d7d82462bf481f4bd87958c9b"}
{"id": "train_037039", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "14470e25565b0c22822ae35c7101b39ef2ef0ad1"}
{"id": "train_037040", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d430880db7f8bbc0362f27b2a4aa6f04cf97613c"}
{"id": "train_037041", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5edca04438bf8059230578cd688e86c72dc5f1b7"}
{"id": "train_037042", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c2641126802ae07e7cfce1ebb62670d3808f9da5"}
{"id": "train_037043", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "0904302c334c8a49742814df697f3de0d6e09764"}
{"id": "train_037044", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "1f368e8c550742776f487286f177c7c2faed8397"}
{"id": "train_037045", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "375f0f3bf4057e7388d95ec0ad1cfbc740f718ef"}
{"id": "train_037046", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "57e09b52eb58ad9880fb0cc5f0f0dcf20fcfe1d7"}
{"id": "train_037047", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "c2641126802ae07e7cfce1ebb62670d3808f9da5"}
{"id": "train_037048", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6862d7f102f1225beaa089aca9087bbcbdb9756c"}
{"id": "train_037049", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7632ba95a314f52efbf5e6680e35733a70da9f9d"}
{"id": "train_037050", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ca8559d05b85bc5636178fb7fed6b51440e181b8"}
{"id": "train_037051", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "26926dbed6352d1f4fa2afb69c1d359956476504"}
{"id": "train_037052", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "6c98377dc6445bb3063a85b44a19ce8db1bc9ee7"}
{"id": "train_037053", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "7e26da5a7edfe8ace3bb98a5cb922bcb44ea6ab4"}
{"id": "train_037054", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "86d592853b631faa2ca39fdc26339c86f475ad30"}
{"id": "train_037055", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["research", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "8bc41e84b3de6b137c1e41454bfd43cf91b1e1af"}
{"id": "train_037056", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "1d0b1034244752c7cd11d1e92d65e13fcd5a5bd9"}
{"id": "train_037057", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "873647e86958ffb2dc68155365f8a3efa20f42c8"}
{"id": "train_037058", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8cc9fabb7dc77b6497bcaec029cb5ae00f842587"}
{"id": "train_037059", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "7e837902a8adefe498082c2b32d4a11c78232287"}
{"id": "train_037060", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "80853098eef4c055e13924d08f1f07fb32d47872"}
{"id": "train_037061", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "97457669f113d89d72258496307a1254be4c1845"}
{"id": "train_037062", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "34ad4a0d51a1183645377214a9449c1db5fb279b"}
{"id": "train_037063", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "01379a56621359229936356f3d341bdb1a0567fe"}
{"id": "train_037064", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ae9bc1e09d7db8d88daef5226f4d9bca11276f5c"}
{"id": "train_037065", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "91125ae2022236376df9a92a10c927b79b6379be"}
{"id": "train_037066", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "36ecead498e21207f73c7ac35fab5bf59005bd1f"}
{"id": "train_037067", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a0dbe8b5f15b312314e873237f78c87edf095648"}
{"id": "train_037068", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ce9f21ab81a3e1f1bf7cd25f9c6e5d87c03ace5f"}
{"id": "train_037069", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f9d2a9103ac9b8fa29ece6659fe001f3728c8748"}
{"id": "train_037070", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "ff9349af12eef5a5f68f292439258b96273a164a"}
{"id": "train_037071", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "15fcc40fffb56450f794f3dcf43a9ea486cebe8b"}
{"id": "train_037072", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "788d1c98ec6b3b91d0ee22494a9ec2264208a7a7"}
{"id": "train_037073", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "a5637f7007eea223d5ee33ac2bfb46757ce0e691"}
{"id": "train_037074", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3f372768e95c7ad6b286b6c803d234bedc782072"}
{"id": "train_037075", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "53c92a3763f79237eaa7eec122dbe0d9af72ed10"}
{"id": "train_037076", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "4c75e54212bf9ff7ed670864925f54e2a272bd6e"}
{"id": "train_037077", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d52d73df8076eea629e68901cb8bc87c4fc99b21"}
{"id": "train_037078", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "120a39301c28c0bc1329751933706e81034d6857"}
{"id": "train_037079", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "4e341022d381dea5ec1e219e390459b39e02ebb9"}
{"id": "train_037080", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "030a82835f3535a12d1e444c369db84da74c52f1"}
{"id": "train_037081", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "c32f3afdbf7bcc6f237c9af7ae36415fc7033cb3"}
{"id": "train_037082", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "dcfa9c92bfa9be6e92b4c1c9d5c0d78095457502"}
{"id": "train_037083", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "06415bdc4cdc81f8cc40119e7a9d747b087615f4"}
{"id": "train_037084", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "49ac8d318fb4e5f672c5f198e43630da3c2ecf5b"}
{"id": "train_037085", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "cc6a407659f946071eac014d5c4692291b2db556"}
{"id": "train_037086", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "fd7fe7d29bb1c1d98ba5cdb01cb3a2f10787d5b1"}
{"id": "train_037087", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "cc51b8bc1539e4a7a9668b824cbe452b5eda378d"}
{"id": "train_037088", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "16896c5f4649459f64369002199a75d38e87bb8f"}
{"id": "train_037089", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "ce4676e413b6e7226ae913b281fcb51b6fc05315"}
{"id": "train_037090", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "015748858be2e2328336c4504bd9588e43a9cdd9"}
{"id": "train_037091", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "09e366834423600bcc06e29004d2299a3b5d0d2f"}
{"id": "train_037092", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "e62e0c878567f7f90c2ad9e4856c99f22a60e8c4"}
{"id": "train_037093", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "eae4f490e8252baaba37a26a6c6e6bf8642fbc9d"}
{"id": "train_037094", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "618ad0ee1875d8df538bed5c8e474f8ad874a053"}
{"id": "train_037095", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e6850e89a2cfdca98a009d2205bf583ceb6140ab"}
{"id": "train_037096", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "888ed8e6b913f94a6e5a751dc7ea597d5a7df44d"}
{"id": "train_037097", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "ef3905575be62dd4a7496d808d1ef81617c3d3e2"}
{"id": "train_037098", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "21e822a54ce67cf72d0a75703b32a13eb3cf4912"}
{"id": "train_037099", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d5ae096e703c4681511e872f7303b3e915e70740"}
{"id": "train_037100", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d56920b4b6b22cbc4aba993d249e883f242ebb61"}
{"id": "train_037101", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "eb9b64cc020624cd683bebcb47cbefab70bcae8d"}
{"id": "train_037102", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6c95eddad30b9c6392b7902dee26e26f3e729880"}
{"id": "train_037103", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c32bfa5002cb0f3f5c2b64f67099e93e23ff1274"}
{"id": "train_037104", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "bb7f2df13db407c5f7646da98304d7b38ab41def"}
{"id": "train_037105", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "9bb75fe6021f02266f1aec5763acb97894d0f286"}
{"id": "train_037106", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "38e76779466257e1c7b0dd04e5c77cba47700ab5"}
{"id": "train_037107", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "7c07146d0c39c737f9401b285ff9ab84e93402a4"}
{"id": "train_037108", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "980acdfc84b042a5d40024ad17319976a4f57f51"}
{"id": "train_037109", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "77739249849b386f6bb9f43b5b511ba15af2bc00"}
{"id": "train_037110", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "1e1e274cefeaa9cc9fc6907eecba15073c118102"}
{"id": "train_037111", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "130dd659e277443f67f67a23656db897a93b7834"}
{"id": "train_037112", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "cb045b3f32eb71efd938b3db3d001ff6067b2b9c"}
{"id": "train_037113", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "bb2bbd290de73dc1114e5cb119e3856e6a515c69"}
{"id": "train_037114", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "8c60bb6a907b93c8b7be7aa7373db9f9283fce27"}
{"id": "train_037115", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "89709c1afb1d85db53f088dd5361634adf11f42f"}
{"id": "train_037116", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "74b7766143b188da1b2053f36f37f514681c3daf"}
{"id": "train_037117", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "4ab057e5ff30e5b9faf743eded7eea42c88c71f2"}
{"id": "train_037118", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "dc1d88eb16ad6c136602191e7352b092ddac55e5"}
{"id": "train_037119", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "39fc14bb3d94d09d053bef941f346da6bf5f5111"}
{"id": "train_037120", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "a1bee81fa3e7ef6171e01b366facaf8c10d521d9"}
{"id": "train_037121", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "15443f7e17a3d51373d025afb51a48eff1e53e4a"}
{"id": "train_037122", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c4639f0eab90e303cd908f95b975a8f1744f0593"}
{"id": "train_037123", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c264fa2091438a81262eedd27aea7fcc49b70055"}
{"id": "train_037124", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "398ecd3dcb19ac1cc6bd8c045ec3bd1c2935f6c7"}
{"id": "train_037125", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "ac2bbfdb44af3f03c1f7a6c46515f17249eb778e"}
{"id": "train_037126", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ac2bbfdb44af3f03c1f7a6c46515f17249eb778e"}
{"id": "train_037127", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "706cd1b7693a20359cca1a0260ffd137d63144aa"}
{"id": "train_037128", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "5f14db28b948f90f67f06e80682c1bbd45713077"}
{"id": "train_037129", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "ee28b0627b9d28a1ce4bbaf4901a57f9e304554d"}
{"id": "train_037130", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "cbabb68f3e5b956eb0cfd7f76507bbe96d57d9cd"}
{"id": "train_037131", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a730905bc81e54ece1c8902705f5e1d9fbd21f4b"}
{"id": "train_037132", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "c71a540bf385e4c3c56a5106bbc7ce9edce16694"}
{"id": "train_037133", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "82b03c12f982747cbc1e2f74ead3fe0b9ffc4e45"}
{"id": "train_037134", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "3a07ec862da4550e9a9bd83a18e2a09c14fe8a34"}
{"id": "train_037135", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "93237fccaf506dfb8346c68544fd5fdfa69d48a9"}
{"id": "train_037136", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "ab42e0d4ee17095bd8ede93e14b8df8f71f976e8"}
{"id": "train_037137", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ffb8509f7ed2b9df601866db9cf50ab421f32853"}
{"id": "train_037138", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3b2d2e8fe2a42346615994ca88763fa35f8549f0"}
{"id": "train_037139", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "39bc78d08fc014a036c46061b994f4aa7c419eaf"}
{"id": "train_037140", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "49041da8f311b49dc4654339926b0a597acbc838"}
{"id": "train_037141", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "b7278e5e3d45ac465da6d77b955706505da0cb95"}
{"id": "train_037142", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["governance", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "1bc2ca8386bd300096bba10f11a1f480ed24dfc1"}
{"id": "train_037143", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b6148328ab4fcf577c80cbd8eb1aea9b4e89ee9d"}
{"id": "train_037144", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "dac2095fe13969fa85239b8740e8ffc1fd905b68"}
{"id": "train_037145", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "08d294869302c0d99bdeeb0d3ccd5399bf416fe8"}
{"id": "train_037146", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "68c5cf14a52ba1287507403fdd0fc8aeec2fb342"}
{"id": "train_037147", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "ac85fd65dd8b20a56cb1f8fb0c86b6007e1a86a6"}
{"id": "train_037148", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7c94b71c332177f88230ddf0657a6332283a5462"}
{"id": "train_037149", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "19630d7124f7403435bd6b172e8bdb24366383eb"}
{"id": "train_037150", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "f264bae0a39cef504b31915d9871615e76755f48"}
{"id": "train_037151", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "af100504f98161778031f87b4d767671dee40db7"}
{"id": "train_037152", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "626cb477fac7e47a44e52258cee5b9d808e78e62"}
{"id": "train_037153", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "737b25f41e45e68d32a5659c05273a0facc5f42b"}
{"id": "train_037154", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d3b9dcf3b69f3f3b555af90f04f68d0c6b270eab"}
{"id": "train_037155", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "5634594a5900bf7bdebad0884bf05041cd886f48"}
{"id": "train_037156", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "5bbc5bd10640d01d75cb386f49c51902788ab640"}
{"id": "train_037157", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ce317bac6d0ecd696b08b6ec0da19be5aefa2b3f"}
{"id": "train_037158", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5d956da6966f6052725d8190a0d2c63f40b58c6c"}
{"id": "train_037159", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "210a2ed9b844bdf8e99500cf402b7aff90684513"}
{"id": "train_037160", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ffad3061c20e95198f0cf38f84b7a83f1523711c"}
{"id": "train_037161", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a9eb6b0a6514bd77556f26109d07bc6981950cba"}
{"id": "train_037162", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "cf02f5cb4cc6a04ff5b7c34cb4c08aeab7174a55"}
{"id": "train_037163", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "208db9e38d0d193edf045a86d25171369a44d56c"}
{"id": "train_037164", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "22333c25a1ee1657dea6fb81083e093900b4ed7b"}
{"id": "train_037165", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "03f5e75dcb663df4a01e4c8faec28c7fe307bf34"}
{"id": "train_037166", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f071ff87e22659f730329f0f93541a25f40a5a50"}
{"id": "train_037167", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "4d630552eb4eb359de5dcc7f8ae1bd4bfa353a23"}
{"id": "train_037168", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c6604c18c2b782e1486bd3d801dcad5303822774"}
{"id": "train_037169", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "dc53f4aeffc2ea27fe26a643f2e4d986febd7ff6"}
{"id": "train_037170", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "72b9c829e85667736fc83ae9f40e72130be6d708"}
{"id": "train_037171", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "80445f62c8efc880c79f13405dea871abe9ca5d0"}
{"id": "train_037172", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "0363ef714f9a2691f614b40e09925d3778a9df6e"}
{"id": "train_037173", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f9a0404d6ccecd63d6a0304ea2c02b0173f57911"}
{"id": "train_037174", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "70385b2087c8091f2cc06b4eee3c209eba0a50c2"}
{"id": "train_037175", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ce2025e57e73671fccb9fda37429d72e0ca97c82"}
{"id": "train_037176", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "38ee0bfc97452c605701ab0a69b6c4b3578c080c"}
{"id": "train_037177", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "7f0ab794e58fb04df248819d2c16ea6134558146"}
{"id": "train_037178", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2ffdf2d1c11df56ecc6a51d009a131c24da9153e"}
{"id": "train_037179", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "8cd44c95d2ce733838e250dfd2eab9306a02a159"}
{"id": "train_037180", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "1d66e0b2c7eaf4c163b03746362a27d92ea1b047"}
{"id": "train_037181", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "325d95724b1c63cdafc67ae296ac43f59ac21f03"}
{"id": "train_037182", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "b9456ef7145462904fcbbfac722374ff82adbe85"}
{"id": "train_037183", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "557c98bd781ace970515511eda4e3c91ef5f2718"}
{"id": "train_037184", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "67cfafe62fab2fa22675ffc2eb2dc3f0c5c1628c"}
{"id": "train_037185", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "dd81866abae578b8211b59792f7529bb90e17ad9"}
{"id": "train_037186", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5fbd8fea0efb2ecbf448efa9fe18460951bfb91a"}
{"id": "train_037187", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "1fd78519679ef08bd4bf863396ac7eabffc09be1"}
{"id": "train_037188", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "588294c820470349cc13e91be0b756f5bf782093"}
{"id": "train_037189", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "38193ef683e303d6346f748a6dc2adfd67de8bf4"}
{"id": "train_037190", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "2a372a13bf0bc4b08fa6a9bfb510c1cbb10b6a2f"}
{"id": "train_037191", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "fc49fa1152085e1f4c9c4f8b2dca0dfcad593052"}
{"id": "train_037192", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "5d5ff02b112401e35111ffeb095c41376fecdb71"}
{"id": "train_037193", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "472be5726038085618f029d1d75239a90ef42515"}
{"id": "train_037194", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "31831699f3dcd9c121b4be6d73cae221ecbf3e6d"}
{"id": "train_037195", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a6d3c7ddebd6be396f9d4c8473f15e00b5b12d9e"}
{"id": "train_037196", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "54264bf7a35c2dc79269601a5f3677fb8bad6250"}
{"id": "train_037197", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "8c5b1bc2fbcdfc51dcea89d0818423d4bd89775c"}
{"id": "train_037198", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "da8bcc1883d19a730360ec3df08e3c71b2cff5c0"}
{"id": "train_037199", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "8f34dad8f56328a7d5f958f0f5ae88979d241158"}
{"id": "train_037200", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "42a909985e0486896b64b7f61ae17f99acfca0a7"}
{"id": "train_037201", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e642f7f16b58f972d97e9da6e6c9b24650042c30"}
{"id": "train_037202", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "75ae19837d9bd71da2c4be66a7e11085f76136ee"}
{"id": "train_037203", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d648a03234bd8468ab72e3d5ec932f07a8887e56"}
{"id": "train_037204", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "a26d367c085e0c10a7cfe4dcf015282567b1ee21"}
{"id": "train_037205", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "8ca0f0e9e46eabb3e6960263e7c1015253c78ba9"}
{"id": "train_037206", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "41b3692a5c024eb40cc1774d945503f54f4740cc"}
{"id": "train_037207", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "2c0f77d45d37a8450a191e3e3d0e43be43536a3d"}
{"id": "train_037208", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "0df3a46146a76a151eb645529390292d99a40c26"}
{"id": "train_037209", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "efef004a05dddf7b2eb1232c193e9af53cb953ef"}
{"id": "train_037210", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "db1ec076cf04f93f54cf303f3b0bdbb0d5e8360d"}
{"id": "train_037211", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "1ffc0b99889a2cbe808fd25582995e68b5ba2b7c"}
{"id": "train_037212", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "25d90f6ba63dabf6eb44d55e836ef6281e09f1c4"}
{"id": "train_037213", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "bd02296d19af3d2c09b447bdd79f67ceea15001f"}
{"id": "train_037214", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "fa680f76e52bfe2d80ccc0a0cbe973fa30a43a21"}
{"id": "train_037215", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "90759ac86be26b6802a420206b9bc6880f60716a"}
{"id": "train_037216", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c9c7e4ea8d223fc373a8b6bd8f542578111bc465"}
{"id": "train_037217", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "26ccc06e63a077472482b0d39e2ee4289eedb19b"}
{"id": "train_037218", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b198047b54f0cd663c82625662119091be871a2e"}
{"id": "train_037219", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b473ee2b911add6c4d262f69780c5b1231319809"}
{"id": "train_037220", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "874fd218fd3d5dd9b95bd30bb3ac6922f2241393"}
{"id": "train_037221", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "e44cc4a3bf0f8595a0535804bbc95b0b5d42097d"}
{"id": "train_037222", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f9b3258c5b5c7988fcadf2571d74f1e93738f848"}
{"id": "train_037223", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "524b0d0e497d0f95b6ff534e164c7e658b8ac428"}
{"id": "train_037224", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "649644c63dd9bc3e852087e40e8418263a3ce308"}
{"id": "train_037225", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "c9c91261cb067ed30a4107a7faff75b229e45a44"}
{"id": "train_037226", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "5cb2e0c2a60daf0c8bab6ddcb40cbf2ac22522a1"}
{"id": "train_037227", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "472faf0d2c62fcd0dc81301d8591bb999a399f66"}
{"id": "train_037228", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "3875ac564e48832891544e3efe68a2acfbd14cdf"}
{"id": "train_037229", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "99c25c2fd9951aae6eccc56b0a5d045c195cef9c"}
{"id": "train_037230", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "bbb76155c1d52f920805a381ff906096f7937fef"}
{"id": "train_037231", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7236449ebe53b096eae2dd5a916818edb5ee05e0"}
{"id": "train_037232", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "174120f7c063836b65ccf5545521a7d5c924bc3d"}
{"id": "train_037233", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "91b51738183f1022d4dd902317752a9179966705"}
{"id": "train_037234", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "80d1eeca04f46d47913c53eb91f23d5d7376a7fe"}
{"id": "train_037235", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "9a9501cfe33e56c5ebe5aca013e48b1d2755ede6"}
{"id": "train_037236", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b5b97af04c6b2d33bfd95217f16f50e8adb5be86"}
{"id": "train_037237", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a084deca0468794fe11e04c0ec15788dcf99e40d"}
{"id": "train_037238", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "29dff1131862c16217bc399a1c409387c46d8dbd"}
{"id": "train_037239", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "5eff43b45fd88a16aead312bbf1355a1bf05c7ce"}
{"id": "train_037240", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "477c1175a48d7d641dc12706bcae5cc6f437249b"}
{"id": "train_037241", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "4e96a277ec2d0f18eef305109e6ef5c0d749b83c"}
{"id": "train_037242", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "2970119bf76d4fb7d4a38b4faeb026f50338f440"}
{"id": "train_037243", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "346b82f194bd17ccc37d899f0903e1cd283362f0"}
{"id": "train_037244", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "45b846bc35f46db503fbceeeff1d35f9ca42cab3"}
{"id": "train_037245", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6b077c53eb58ca22de3382578910911d195a3263"}
{"id": "train_037246", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "fdbde1e1f8923508b301e49b8cd52894e503aa39"}
{"id": "train_037247", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "dc29ef4e6ab311b1812cd064024aea43447b96d2"}
{"id": "train_037248", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6fb29b368e1db895dd8abd372ccbe039d3e5613b"}
{"id": "train_037249", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "8e4e4da46b53256e679897a74aab6b2910a7984a"}
{"id": "train_037250", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "72c3e5e37df11a2850d034fd6663e6059d00d26a"}
{"id": "train_037251", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "759cbf3afbd3a2a6fda80f76ba148091418d3c31"}
{"id": "train_037252", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4c46906515bc756c48a8656dc4ddcd5b8318ba5e"}
{"id": "train_037253", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "3bd4d365b62cfe698081b2aeb42c34226e3bf620"}
{"id": "train_037254", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["data", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2a7d7e2e097419834d7697c4f61049a24feaeeba"}
{"id": "train_037255", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "ac994b6df4e88b13a0c0e07c01af058f3a3beffb"}
{"id": "train_037256", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "da8b38d463fe6fd33a480b22a113143adde8d7ba"}
{"id": "train_037257", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "1814a4724b63fad159dbcba5c14d14c7b5f2567e"}
{"id": "train_037258", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "12b5ec1c6aa2b7dc9d2c0db68cf8f0ed5ed2d689"}
{"id": "train_037259", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "1917cb83cbdeb61c3985c116dde2655a08fe14a7"}
{"id": "train_037260", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4af3b5e3d91a154877c7b41c25b304fafe5a7121"}
{"id": "train_037261", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "fc8172e5aba6ff1afa58559d91bddcc4e1e60802"}
{"id": "train_037262", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b7b8f0372f990b8941209012a03a96776d1139b7"}
{"id": "train_037263", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6289e46c9b6f66ce5b0db70d4f4c751b94f48eb2"}
{"id": "train_037264", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "88d9d4b992202bb678b207794c29ba65b01fcfdf"}
{"id": "train_037265", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8d83ad86ef8ce764ea9af59f3a36e94792e21e91"}
{"id": "train_037266", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "09d489428d2ab12b9fba0f08f79bdac0daf863db"}
{"id": "train_037267", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "9ec7f5a9874e7dc853dc4a3bb0c6daf39fa1ebcd"}
{"id": "train_037268", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "da835465794a8cccf6efbeb0b24b2d62da833601"}
{"id": "train_037269", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e514526fd3c212e60d87fd659808637e20c0f4ff"}
{"id": "train_037270", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a45446a4ad3df2fdc2f42992de3c36016aa235dd"}
{"id": "train_037271", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "081aa17079ec748eaf4f5ca708f9e536daea2ae2"}
{"id": "train_037272", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d1419f42b8d2edb800de4c9944a4dea3413f41d3"}
{"id": "train_037273", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "1add01765b0ba2c61c7c2e782b967fafcebe458d"}
{"id": "train_037274", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "9b34dba91b148c785f3704155a9b00ae484fac9a"}
{"id": "train_037275", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "049690ccfe069c94a1135dd2330558dbfaff75d5"}
{"id": "train_037276", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6f02359e139ef17e66177b22350b8332c27c2396"}
{"id": "train_037277", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["governance", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "388ea4cdfa916291428fcb2013c9d93c7d5e6810"}
{"id": "train_037278", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "33b2fcef729ab07bf7bbe531dd24c49b72af0ab2"}
{"id": "train_037279", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "81e09568a535df0012a1063c626d8e5e5712bb41"}
{"id": "train_037280", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a537ee2978e640c477bd26bee97abc671729c521"}
{"id": "train_037281", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "7c695b37a770efc74ed866dd503e1a079898674c"}
{"id": "train_037282", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "9b07deefa5bf1f82605be0c94183b3d60090933b"}
{"id": "train_037283", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "e83a38ed3aabed5b2a3f673fa35c9206ea390f53"}
{"id": "train_037284", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["performance", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "ee063c1d358ec08d25aad03aabbf6609485460cf"}
{"id": "train_037285", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "809cdf85d6fe167ebdb99e1bd9b0fccff576a5a5"}
{"id": "train_037286", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "3712918e80ed5f05e1aa3517d285cd269e38cf5f"}
{"id": "train_037287", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7e837902a8adefe498082c2b32d4a11c78232287"}
{"id": "train_037288", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2e285b46fa7d14dab725963aea16cc335447541c"}
{"id": "train_037289", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "318859f6ef90de4f80d91fdde08a1b20f57cd5b7"}
{"id": "train_037290", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "9523ec4a9cc6fe6abb415a6384171c93f4bdea1d"}
{"id": "train_037291", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f2bed33c5604472090bdc09764917972d9ed6541"}
{"id": "train_037292", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "85ea21db941e346494ba415a17ac269dea538e15"}
{"id": "train_037293", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "5aed8cea893eac75d0add5c2dd3b7e0b802db8a8"}
{"id": "train_037294", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "dce93b042cfa2518401973623e98b1080781fc8b"}
{"id": "train_037295", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6a8dacebc2a8959494fa2f9154b68b96f9451306"}
{"id": "train_037296", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2aa938ff8f81d7bed2005e17f91f6e92cbae1516"}
{"id": "train_037297", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "8bb2fea9655fb2ea835b190c026540cda7bd61ef"}
{"id": "train_037298", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "52bc9ecbd01a8c04c67ace07a85d7b496436c810"}
{"id": "train_037299", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "c7c15200eb61887ec67019e37bd65e50f813b5d4"}
{"id": "train_037300", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "afa1b36b47519fbf910623732daa454876f6d2b8"}
{"id": "train_037301", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "bad8286510b7240cc198b05e8cdcc96e5d52fd30"}
{"id": "train_037302", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c214041a2d4110da055eace1c6a95f8ef72c1c18"}
{"id": "train_037303", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "e920e28478dc5def356b38c01937afbc5fdf9656"}
{"id": "train_037304", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "034664b2bba7d64ea5881379794af9c27abb39a7"}
{"id": "train_037305", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "51ae8e3de7e07719f80c238111b74e861b24713d"}
{"id": "train_037306", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "44e830bd4e503168f3cb87ba35b6ba114fdefdac"}
{"id": "train_037307", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "3c509973f1fb3e193de9787765d6367f5f4a1680"}
{"id": "train_037308", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "3db46754738bde6cb8d655ccc406c256b30ce0ec"}
{"id": "train_037309", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "be739356633de016669daf35536551732b59bf67"}
{"id": "train_037310", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "02c58c866afab0b9b70edc846d6d4a6c5e641289"}
{"id": "train_037311", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2ea451d71b69e77d1576d0b0a3404572614568cb"}
{"id": "train_037312", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "6eb4523da7f45b5bcb79aec51f1193d8d09899b8"}
{"id": "train_037313", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "3d013d439b8496b8afbd2fecbd1115a8526839e6"}
{"id": "train_037314", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c14ab374b8584a723ee18b082e52b7a6cdb91a2d"}
{"id": "train_037315", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "fd28eb56f0daea90015bca7fa23aa7c81b845a43"}
{"id": "train_037316", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ff1e7cbb3fa5930c194a231c02415b459b6a42fb"}
{"id": "train_037317", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "577e4c33ba94afdba82975f04a85a11612db651a"}
{"id": "train_037318", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "db2b90b72f15cfbac610539843f90f99772aa40d"}
{"id": "train_037319", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "17f6f2235e8ca97ecd55aac6321075fef4117954"}
{"id": "train_037320", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b3842402629c59c952a9cb7853ebb5c1980ba926"}
{"id": "train_037321", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["agent", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "4f86490865ee27e8754264b095b3e4ff9799f9db"}
{"id": "train_037322", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "08203c4312e2c36e9981b6135e45279c75811685"}
{"id": "train_037323", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "dff3cfb9dc0aafef3f15ba0d8c736205cc6d9475"}
{"id": "train_037324", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "8bb2fea9655fb2ea835b190c026540cda7bd61ef"}
{"id": "train_037325", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e44b50090d7a38d2d6b013c78646539f23623b83"}
{"id": "train_037326", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["research", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "75084af31552f3a8e12cca27f803fcf6a23422b0"}
{"id": "train_037327", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "96801fd64cdf90b9beca79cdd96085dbb7233ca1"}
{"id": "train_037328", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6453c8e0ca15843cac1e3d18ffeddeea8842f887"}
{"id": "train_037329", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ae1875a74e9c2cd933835e6a793801d3dc17412a"}
{"id": "train_037330", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4a0c5d2187910ee92b47bda3fd8d5294dc2d3310"}
{"id": "train_037331", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "c4fb3d690efe3905da4cef5456725c9e91168811"}
{"id": "train_037332", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "5210f23153360b52e84176590f9652d9bfd8efdb"}
{"id": "train_037333", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d51a267f329a133c4e5d2c69045b6ddc4659c599"}
{"id": "train_037334", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "5516ed0ea400204216756f398212ca835c56f0db"}
{"id": "train_037335", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "af100504f98161778031f87b4d767671dee40db7"}
{"id": "train_037336", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "43c0c220fb4f9f48f42db0ac05871279e2970590"}
{"id": "train_037337", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "65ddde9ac07963ca551be9a5c6b9ea906bcedf6e"}
{"id": "train_037338", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b75d3acd5a8fe167006098655e7fb38e29e38681"}
{"id": "train_037339", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "4a7fd23982bedb755e9b79b89aae38de9404aefa"}
{"id": "train_037340", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "178548142c851fc6febb65768198d94b6d69fc87"}
{"id": "train_037341", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "838ee001175ba17ee5c59a71cebf086dea240406"}
{"id": "train_037342", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f2499a642c9ff7a52f0e238dfbfdf86167610209"}
{"id": "train_037343", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "bdfdd2bcb8b3ce897c0a5a3aec2f42ae0480aa30"}
{"id": "train_037344", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "63f7e45f705868ef7625cc6c25c4620ff74a8e18"}
{"id": "train_037345", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c89774f32a51dd2b23b397dd6cdc021557a0e5ff"}
{"id": "train_037346", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "21b0d2f0d758e98cffbf685ecc147c0c0fa949f1"}
{"id": "train_037347", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "44763ac94da2b4d8ad7dc4b5aade54dca73d2c6b"}
{"id": "train_037348", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "1f2df978b9b9d87f488616df73d1c4a94352449e"}
{"id": "train_037349", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "da9b942637130ed1d3953cb6b076279c00f8282e"}
{"id": "train_037350", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "57a7cab42f6a2ac6635e6b4710fb59693c62c1de"}
{"id": "train_037351", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "365124f94d13dcfbc13f7cb2f4385eb68a09bdf3"}
{"id": "train_037352", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "687e213332f8e21d5728f216cd6d74870884d0c2"}
{"id": "train_037353", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "a4dbc7384d360eb86611d5787f6b133a87279497"}
{"id": "train_037354", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6bf9d9d691d34898a2baf93db837ef644bbcf355"}
{"id": "train_037355", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "55ac8e868de881c5f4e444008b58ca2a2c053aa8"}
{"id": "train_037356", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "3d466372c67a8210216d71ee576239c925ddc60c"}
{"id": "train_037357", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6454c74b02acd946388ce583976c51a3421fdc80"}
{"id": "train_037358", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "027ff281d15402b3bcb2e762ef4969d7722a6e2e"}
{"id": "train_037359", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f6333516e0a8557c13485eb8e720fe80a82aa7c6"}
{"id": "train_037360", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ac922b6edd9a59c3e137194e6c6758ecf18f4647"}
{"id": "train_037361", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "e3d00bca8b0b29402d7808672c8cbfc9f50bd1a8"}
{"id": "train_037362", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "ad4cf6b46dd3156b4fd1311b83cdea1b41cc257d"}
{"id": "train_037363", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "9e17ea2c49f3092c07a970cb1d70db5310921734"}
{"id": "train_037364", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "80d1eeca04f46d47913c53eb91f23d5d7376a7fe"}
{"id": "train_037365", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "b1a6680afa307e064f7be1a8056b3d57b15d45d9"}
{"id": "train_037366", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "83936dd2c96d048ea38493459b2dcc25d4da1fdb"}
{"id": "train_037367", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c89774f32a51dd2b23b397dd6cdc021557a0e5ff"}
{"id": "train_037368", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "06966bd304fa1a508b00dc89b39a358242d2810b"}
{"id": "train_037369", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "92304c8619e170b4061e5fac0309e6251b3aac0f"}
{"id": "train_037370", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d7bf57fd2144e73a9da2c3dc242e758ce04e1de8"}
{"id": "train_037371", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "973cee298900f43bb58d1fc0b1c84aa0fe9f7fd8"}
{"id": "train_037372", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8a33e429b0817c7bda9d3061aca23c8668a6db40"}
{"id": "train_037373", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6edc00799b28e0ca20c570020a4f93f855658bf4"}
{"id": "train_037374", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "377348779562821db19aa11f15d2c0727c413898"}
{"id": "train_037375", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "4ced41f7f79c7a74f8ff456125972fe6d2a30a80"}
{"id": "train_037376", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "9e8921a3e64c41cd8ebd58583ffa4386376dd658"}
{"id": "train_037377", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "5795d7df9ac6c8fb5b4bf8e51d86fef7c90eb5c7"}
{"id": "train_037378", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3614b71a194f1862537dc0ccf14fe9125083201f"}
{"id": "train_037379", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "be1add725758dac9d1caba8ced058b52aee3cd85"}
{"id": "train_037380", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "f20550c8d686cae2205376f9eb82bfa80f466e0d"}
{"id": "train_037381", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b02888d2fb2af5bd5eac9502908d110cb2111efa"}
{"id": "train_037382", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "5b0b84538184c6170515ab576f703a4fa27afc11"}
{"id": "train_037383", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "009c86852e0325875b9ca8cd17c849e540160c7f"}
{"id": "train_037384", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "49d0d73687235b9d0e51dc736c071ca0ac1b08fc"}
{"id": "train_037385", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e5cf94ca38bda2c2eb5de903f8f2f6b55eaa5146"}
{"id": "train_037386", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "6a0239bc8b675fdc37d8022d1a8aa2dd25426267"}
{"id": "train_037387", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "1029e6b7b6015a72113d3e552368621f23a48c22"}
{"id": "train_037388", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e222969d95e6a028e2fcacc3998ffc19873d1b93"}
{"id": "train_037389", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "67581b9c2389174c49ce5d5f96510d705289aa34"}
{"id": "train_037390", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "810ae59b6bf71b5676df6a7fa5267bb90c71d6f1"}
{"id": "train_037391", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "af800bcc9d57cefc93c725f95d44571b741a9105"}
{"id": "train_037392", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "4306a4389ca7691c922d3a15cb82233e1be650de"}
{"id": "train_037393", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "ae4c1a26d4676010b3484ed9b5c92b9ab33bdfc3"}
{"id": "train_037394", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "07553415fa80a7a82e1053504056d6eb1bea7d05"}
{"id": "train_037395", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "bd36b01f59ba9bfa6884a32b4d399b8b9384cc70"}
{"id": "train_037396", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b3c7564f745b9cd69cbb72c290aee29c8675f996"}
{"id": "train_037397", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6d2d74ae1cd0237fba4111a9b4e4aaee4be59734"}
{"id": "train_037398", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "aaba2fbce369d579210c09d888d226e355c4ba23"}
{"id": "train_037399", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "641cff9996fb4ab584b8b60c7aec518e746ed5be"}
{"id": "train_037400", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "42be3ad7e8ccc428999ca57c195740fcb3c83229"}
{"id": "train_037401", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "886aec2d53f980d0954137445c6e6ceec2299c38"}
{"id": "train_037402", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "36d1afb3beb8d5a5f273fdb10be6d38898b1143d"}
{"id": "train_037403", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "df2a9d4a415d3e022001b9ec1241b11e55d0f015"}
{"id": "train_037404", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["performance", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "35f7f3fbbe3065ea55fad45e40545d53cea0eead"}
{"id": "train_037405", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "50cf7b639eeaf57fde2577f2d38407e144560370"}
{"id": "train_037406", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "cd644527648db31004ce1cba5067dc2e36b6ff8e"}
{"id": "train_037407", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "db361ef9a69f7ffbc0b43319dd1ba297bfa7471c"}
{"id": "train_037408", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4c1ad3a6bff79287f71c27c9b2b44c8885a7c51b"}
{"id": "train_037409", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "07fa16d651b9914161606da62a86c6e54bce50ff"}
{"id": "train_037410", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "61c772948ae80f6c60d401bd5d45789d0c391a66"}
{"id": "train_037411", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "e7439bb1dc7b1e3722948496c45a04c9800c1187"}
{"id": "train_037412", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "50abb211c14396e8d3ee5d8f0da1bf9dadf904b3"}
{"id": "train_037413", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "0908fca5872a030e5e1bea7cbebb327e940bde20"}
{"id": "train_037414", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "56e28cdca4dcf5b93991756be27deafa84034f4e"}
{"id": "train_037415", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e703207a4b8e6178c178c801d78ee0b43425b7ae"}
{"id": "train_037416", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "df549922e062460e57d49e5e316be49a24609f15"}
{"id": "train_037417", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "3775dbf9d8880a1492dce4ef57bf5cfedc9e3c4b"}
{"id": "train_037418", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7e20fdb29d136349466ca4a2a19127326e9d0263"}
{"id": "train_037419", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "a388425dc7efaf00b2d9c99ec2df760773074491"}
{"id": "train_037420", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "5c710a5456507c4a5c1c2a682575087600d9c4e6"}
{"id": "train_037421", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "3f12152ad5bd1f4146b4c1c83bb4b51361fb153c"}
{"id": "train_037422", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "b527bd368aa106a8320746f63f06f4427a4e1ec3"}
{"id": "train_037423", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "089550801e2917348bf0d0ade532002673f5a0c6"}
{"id": "train_037424", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "25bf353a755acbdf53fc85de492aa8104862af69"}
{"id": "train_037425", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "4ff31c7237d2391171777f7863a0dd1ae37ca57f"}
{"id": "train_037426", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "78638b8ed1081a45332eb08ed399912fcfa258c9"}
{"id": "train_037427", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6b922d0b62c9b19fa410122657de5b928e8e106b"}
{"id": "train_037428", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "5705ef3eb4ad3a5941460535e813e238c10fa9a5"}
{"id": "train_037429", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "945994d2429904a0cea06fb02aa26a6a49ccbd70"}
{"id": "train_037430", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "942453d70527173e2bd62dec865e6b58abb316b3"}
{"id": "train_037431", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c26ac81d188b5f5cbd53349cc46f1bdc26cc1b21"}
{"id": "train_037432", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "56cd19642a5b83690b045a5890b7d4f139d6ef47"}
{"id": "train_037433", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "bddaab2c7d8e6be63bdbd16d2eee468ad1e72954"}
{"id": "train_037434", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "ee8f3b6b0907810d0cfba266273dd3f09fc4e511"}
{"id": "train_037435", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2f3f97958568ee58b245dbe80e7a5a60d8734347"}
{"id": "train_037436", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "f1af1cff96f9bef9c278709214ddbde817b4bb1d"}
{"id": "train_037437", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "8316fa8633082ec7dbbf8a299fb6506c6d71d419"}
{"id": "train_037438", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "87cf252036264b3752f7120b0934a1f9050fbd97"}
{"id": "train_037439", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "90840904bf57dcd5b9c62ca6459816a90a1a859b"}
{"id": "train_037440", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "9a3fa15cad0888d8b3a180ba31bfb2a3dc22d435"}
{"id": "train_037441", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5779f28722042072506c7d9dfbf11b76c917a089"}
{"id": "train_037442", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "fe86ec164457fa111239d49e79b54eeab81d46e9"}
{"id": "train_037443", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "29e46f9a233e53b81e021d6809056d1e58994734"}
{"id": "train_037444", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "d2a89726f7ac317902cdb8b811a1f5741e7bf12c"}
{"id": "train_037445", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "3b68b41d5c15b0a13b248c9d8eabbd52383ea6a0"}
{"id": "train_037446", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "c1f47e7f74c58828dd31f1aa965fed595dee80c2"}
{"id": "train_037447", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "9a68fb6cf24379300b0ee071646420a08734a00e"}
{"id": "train_037448", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "fc639d8016ad5814be93d1dbca93de8113658519"}
{"id": "train_037449", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d9242ee975f7bea792ba1ddaf4821921b1d8535b"}
{"id": "train_037450", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "dde118ad6799a5a41839e91d19578c48c41f7a68"}
{"id": "train_037451", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f0e4d2e32be86a4005d8aa3aba3b183145034df2"}
{"id": "train_037452", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "290da7e2caadd9c61617c64348a3d0b280715e32"}
{"id": "train_037453", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "1a48e5b42c0d3ef72e2a4664497497aadc8d821c"}
{"id": "train_037454", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d7bc156642229da6d4cb400a6fe582bc435431d2"}
{"id": "train_037455", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "2d8a27234a8488b4d984da7e061bdb8fd07bf8e5"}
{"id": "train_037456", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e8c59c7e5e824de0d9ea6255f971776f9e8c3c0c"}
{"id": "train_037457", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c6975f758b727964e0e8c132a2e97e9b250c30d3"}
{"id": "train_037458", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "da8b38d463fe6fd33a480b22a113143adde8d7ba"}
{"id": "train_037459", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "1d517676754365328e1ea5c352e3ad912c404d10"}
{"id": "train_037460", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "1dbbf585be1ecbccdaec16a3d249a2dc32ddcfcf"}
{"id": "train_037461", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "4c03a868799e1b5e7effdb2147b01200a470296c"}
{"id": "train_037462", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "0519fdc859f87e5455039159378d215a95480b1f"}
{"id": "train_037463", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ec5aa7d133507d5b0ecf20602aed588cca1215e4"}
{"id": "train_037464", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2ec6001b62c95241772c67747bd9f4b72e791bfb"}
{"id": "train_037465", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "03f5e75dcb663df4a01e4c8faec28c7fe307bf34"}
{"id": "train_037466", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "a7722fd6b2a3bcd970247eaeedd3ed400c92ef9b"}
{"id": "train_037467", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "9faaea85eb91c55187d66811676ce4e9e391ea5f"}
{"id": "train_037468", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "17e8212aca3f75b9d3ea8ee8b7a3de26fc3ebc21"}
{"id": "train_037469", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "775f8fd200146cc609ef804426fa075e8c0ef4f6"}
{"id": "train_037470", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f4ea7ad165742c25bd6bc17975806331fb231c7e"}
{"id": "train_037471", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "9823ad6ccbdcdcf92f0e2c8f718f61ddaf3555fb"}
{"id": "train_037472", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "373fa3c071a8c5f1167c03bae835186e42b3be09"}
{"id": "train_037473", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "e6e9c9179945e1628cfddd493e4ffc6692e95b52"}
{"id": "train_037474", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "99f3e5c4876fe240e2502cd662b2a360f77d9f3c"}
{"id": "train_037475", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "987a7fad340ce22d3d5ab5b4037116f4da54937e"}
{"id": "train_037476", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "245c29dfdc00fe19cdcc1fa7cdd906ce571fc3c5"}
{"id": "train_037477", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e0a25735124a3c77bb37d05a9934602e7998b460"}
{"id": "train_037478", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "409823b28396a302994eb18626fd68f105c08204"}
{"id": "train_037479", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "38c8fa7f6e865ae5ae4b414cfa7504071617d163"}
{"id": "train_037480", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e6ce236eac5383cd280343aa2ea0bfb5eab1a45f"}
{"id": "train_037481", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "618ad0ee1875d8df538bed5c8e474f8ad874a053"}
{"id": "train_037482", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "3f2daaa9da64bb474faf682efd64efbe3447fa88"}
{"id": "train_037483", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "bdc5826598fa896af658672efeea68e21de27938"}
{"id": "train_037484", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "937a91312495cde9aa5ed36e8590ba03113095be"}
{"id": "train_037485", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "5b3d597eb88cf9320f7de377351e0f4481f54149"}
{"id": "train_037486", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "1a3e1d1035e999600e0057381db6afb6f204b4a0"}
{"id": "train_037487", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f700eddc174448d3d61bc8734d5a38d409f9c6cd"}
{"id": "train_037488", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "aa10bfd5fcc7404890b3d88500d1246e6c8d726f"}
{"id": "train_037489", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f2f398a3e3cbb2b0a67d571ad03732e0b4bc132c"}
{"id": "train_037490", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f7a9453046b3de095a11dc9f6324bfda03084f4a"}
{"id": "train_037491", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "57f396d2fd59080872e49ab32a71bf717f86c76d"}
{"id": "train_037492", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f4ee5b4debc8bd5d6ba94ffd7c5c6366968fd5af"}
{"id": "train_037493", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "b38a4129ba37a6baa2fa175a8becc5512baacc50"}
{"id": "train_037494", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "69a183765015a600a675f0a689738b3aa56ce51f"}
{"id": "train_037495", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c9544fa927ac0ee54fc7088be2c73995f62ed7d4"}
{"id": "train_037496", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "8550fc3199d752d6ab29849bdac9d85dbe250f1f"}
{"id": "train_037497", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "2ebac9e136a595748fc8c8852a941f0f65c694f8"}
{"id": "train_037498", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "2545ccf251a15b4fac1a9737916c9d9e2214308e"}
{"id": "train_037499", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "0064c6dc43c673d7d61bb6eb6a058bd862769539"}
{"id": "train_037500", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6420ab532a94800a133984c3c0b929c0b3d268cc"}
{"id": "train_037501", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "9b1cdcdd4a3a1379755c8524910e59b00915c086"}
{"id": "train_037502", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "ac24661079f2cf416baaec347458b03ee16960fd"}
{"id": "train_037503", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "c7188caa6b8c5ad5df423e0ed41f291393afeaed"}
{"id": "train_037504", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "226c8191b0f59af76ee3790cf246557b9b0b48ab"}
{"id": "train_037505", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "4c75e54212bf9ff7ed670864925f54e2a272bd6e"}
{"id": "train_037506", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "9dd6e4246a82398840e1f794d3312d20d9302b9e"}
{"id": "train_037507", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b86a227932de3a28ca5b1149cde1f524d5ff51e7"}
{"id": "train_037508", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "106a15cb61098d201ae5c98bf4e76f06005de80f"}
{"id": "train_037509", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "0ba8324e520d82fcfa67758769c652109bbd584b"}
{"id": "train_037510", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "cb82c6edf59b800fc49cdc26e065c692777c6560"}
{"id": "train_037511", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "4e485f92a9dc32f84decd65134ba60c448f1b805"}
{"id": "train_037512", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "c9a481ed57525e7b9c9eec428090a7ca056bfc34"}
{"id": "train_037513", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "e927d7a23de86170a7b23b38114e04ab50d8a988"}
{"id": "train_037514", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "5b92fe5cc07355ad7df68a3c3001ab455d0a0b3c"}
{"id": "train_037515", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "4857dce2c7800141f56d329218bf4fd2ec0787dc"}
{"id": "train_037516", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "974317bf967f6e246a7e80eb6a4d07b2d9226bc9"}
{"id": "train_037517", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "45e13dcdcb1d2183ede6f6102d22ff1910844214"}
{"id": "train_037518", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "53e62004cdd7c610d5ca560730d7d378d8663de6"}
{"id": "train_037519", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2d2717510e415ff658e124280320b2cf4d6181a6"}
{"id": "train_037520", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "5756b757b244cef33ecdec1ee8e5adc8f555885e"}
{"id": "train_037521", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "97f51486d463b0a2ff806152dd3914bd2bd665ce"}
{"id": "train_037522", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "5fb0d1d63e591a773ec7ad12fa55b945e625a28a"}
{"id": "train_037523", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "4afe449bf7ca4c4bae99b371550d3de2c7e21481"}
{"id": "train_037524", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "b017bb231680a849c085a7991dd6c72fd45caaa5"}
{"id": "train_037525", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "824ee22dc101327bf2ef05fe52ace6a3c1591af1"}
{"id": "train_037526", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "9a914fb6862dedadc31ba4b4470126068c2fab97"}
{"id": "train_037527", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "9d44b0595ef642771e4f504f4cfae5e05c3079c6"}
{"id": "train_037528", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["data", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8d3855e332e9b440d3a161e80284ae3665468f3b"}
{"id": "train_037529", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "db998cbf63f233158924add6e9fc00aad233a801"}
{"id": "train_037530", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "e299ec84837330835f1af5ba3e69a070fb2ea6c3"}
{"id": "train_037531", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "dd127e934ba4a3d9a8895bebc0f07c59b61368a3"}
{"id": "train_037532", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "1e3e57c272156fb163e6f28e8bae65b4117a96ce"}
{"id": "train_037533", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "627b96c9fe341cd9edd8565de3ad918ed5dbfbd5"}
{"id": "train_037534", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "106190dae60a369bc42dd1d832c658d5512df665"}
{"id": "train_037535", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "a58f1950c4d51a4811aa1686b5e7a958359792d9"}
{"id": "train_037536", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "0c34ce93eb4a0dc797a003d1309bc618b107f343"}
{"id": "train_037537", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6f0b13c0d638788118ab90c92154048c95c566fa"}
{"id": "train_037538", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "03b4d05e85ed0d41d5f2b8582f416ffa2a65147b"}
{"id": "train_037539", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "1ce81d5d76c6eae389a61249ff13453bed298af2"}
{"id": "train_037540", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "919f30984f27cdbe9e0e96c26f881fe29539a24b"}
{"id": "train_037541", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c654d07a6bb59bb730fa4bcf35f27e3e0c6a2bfa"}
{"id": "train_037542", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "e8a421c107679864f613d238614f4f9b5d907266"}
{"id": "train_037543", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "2ef3748285da8790cd5709f6c6e0ccb4745f1349"}
{"id": "train_037544", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "4a5382b8f644ae6c75c6c6fc1a5343342a7c76ef"}
{"id": "train_037545", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "a778921611932b713202311ab0254226840a9a8d"}
{"id": "train_037546", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["agent", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "eb72079b3a772f4e79da612ef4eb2569c2892c8d"}
{"id": "train_037547", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "d8a6a4c4f5d0b99029602747e6b4fc6a8237040e"}
{"id": "train_037548", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3ed6c1fa737a4ae04562da23d465c1b6d178c145"}
{"id": "train_037549", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "0a788d3399f731d27213b613846eadcb267a3a86"}
{"id": "train_037550", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "8e43c34b0f886f261c9900979de66a60c782513b"}
{"id": "train_037551", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "af45330e06236d5d9be8dfe5a34f0c0c699e0acc"}
{"id": "train_037552", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "7a4c9f2e465f584d7c12bdd0caae1e82c7a6b986"}
{"id": "train_037553", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6788595696d12eb3386295def7611e173310ba88"}
{"id": "train_037554", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "7df38f78c8bf6e90ae85375554fcc359659f18ca"}
{"id": "train_037555", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "77b897b2259b7f7acf4560a38c8399b7d4030091"}
{"id": "train_037556", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "91d51f9cd9074963e93834b8fd75e331df164aff"}
{"id": "train_037557", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "3f3263ed2e7781ee1f22c3423944c956c8033dd0"}
{"id": "train_037558", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d9019e54144d513edf30a3668c969d8a0742d20e"}
{"id": "train_037559", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e703207a4b8e6178c178c801d78ee0b43425b7ae"}
{"id": "train_037560", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f4e6e0a01049496bb4f865eb63c2f1d5cf207455"}
{"id": "train_037561", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "a18ef852f425fcad6a654f11f01890fb471d9c25"}
{"id": "train_037562", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "cc7e5d5f30b32bafec946cf0c6057b007a844519"}
{"id": "train_037563", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d7429b28a4fe1b3af141fc4f72a30951522cbffa"}
{"id": "train_037564", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "c0f83b7074e9a99fbc3fefd43d6e2fe3d55c8779"}
{"id": "train_037565", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "feabd844e83b5f3782050f116e6e75af946420ca"}
{"id": "train_037566", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6d48f9391d5823d521b5d254cb72b0f2e55b884d"}
{"id": "train_037567", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "0eb69b4bce7014c22ece61bc595168a841d37ec3"}
{"id": "train_037568", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6a1122c2b97c6372005bd24b97fa2a80b8f56d27"}
{"id": "train_037569", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "a0d794e2373ab7f43458ba4c509c7314984beff2"}
{"id": "train_037570", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "11d23917edc44c540b384b9aea6b669805cdeb0a"}
{"id": "train_037571", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "8d1bce340828771bd292e853ef88fa43d26c40c4"}
{"id": "train_037572", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "69b2914ea6e409476585f7949bf3482d0aa3f188"}
{"id": "train_037573", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8fd7c6c1e87d97fe7b78b599fa4bf2a7d85101f8"}
{"id": "train_037574", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "cd9964289377d62866b020f4afabeaab72a667fa"}
{"id": "train_037575", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "7d1b94ce2eaa1e3c7056db2e1d4148f12b16ce8b"}
{"id": "train_037576", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "7e9e093ec45e5d0bec84022690956075b77648e6"}
{"id": "train_037577", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "940d788056fdb513c5e4d501d877bc6f99dff983"}
{"id": "train_037578", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "95115eb029c076a3e25027ed6f37ed9ffbc1242c"}
{"id": "train_037579", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "005cf6f07123af00845a56567dab8d5f996f1f3d"}
{"id": "train_037580", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e9a26e37bbc6cf5beb7e0d7f81264f32e5361e46"}
{"id": "train_037581", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c57454a4da244067f0c863750e23859003e61124"}
{"id": "train_037582", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "bb6fa806559df2823432a6082eda512b11f67917"}
{"id": "train_037583", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "3af8e123101217df4ae617e4dbbb159c1234b1f2"}
{"id": "train_037584", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f7a487849651b1bffb21b1d725a12f1229d842df"}
{"id": "train_037585", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "9d9f0280f4a75701b5d0bc266909699d6fd5e44a"}
{"id": "train_037586", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "28722df3ce46b86c17fa65f020abaa5c1a4797e3"}
{"id": "train_037587", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "e438b48d180750de3c5ff71f7b1b0143790fb563"}
{"id": "train_037588", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "bf70d86ae605882977360d691a14398489f1ef25"}
{"id": "train_037589", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "4c2f1003f0e0d351b9231d58d1b8f469396f7967"}
{"id": "train_037590", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "85fece1b48f9d4f8334d106d1939474d47e4bc5b"}
{"id": "train_037591", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "a373e65b80acb695a12ae8427e568bd3c7280f3d"}
{"id": "train_037592", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "616085785a0d15d86890b50429d554629e2ba1e6"}
{"id": "train_037593", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "319a6ae90ca1ebcc56a7a7b715bc2c90bfac7f67"}
{"id": "train_037594", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "4c960e2b387c35c26eff82016a33f796680c2b85"}
{"id": "train_037595", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c2efb6a0d1d1c7f888908e0e02cb996c89dc29b1"}
{"id": "train_037596", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c5fa512ff6fb50f1cad46f9483c03c7963991ac8"}
{"id": "train_037597", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "710879a07e2e5af5f6b029d4285898e3f6267423"}
{"id": "train_037598", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "01252c19fb9d06c0669971b0c6d6561512783433"}
{"id": "train_037599", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "0d7d9386b2175b90174ecfd435b0f5492d0cea1e"}
{"id": "train_037600", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "60f3fa6840b5cdb974609b8c1bf3859475bd6a7b"}
{"id": "train_037601", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "246682804746e6854857cf445943b0925e062c39"}
{"id": "train_037602", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "33330e1315bd450b9757849902cc365375bcdbdf"}
{"id": "train_037603", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "d39ceaa4ca505cf43b29e68429b2e2d8c8619275"}
{"id": "train_037604", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f6423adcd70bb7f5406407f3e9e6cee39f3b929c"}
{"id": "train_037605", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "3ac648631d39f9b6278e82a1c8d200f3cb906dc5"}
{"id": "train_037606", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "184ba9fad9b1029390b61bc67aa92036137c0f3d"}
{"id": "train_037607", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "69c624b1e45658cb04eb8801d613386d72be7683"}
{"id": "train_037608", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f755910e3515df5e8193acdebdb8b04da1f3e48d"}
{"id": "train_037609", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "42bd266260b033157c1aa179b18e5774f7b8639b"}
{"id": "train_037610", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e99abfd56d5375574c1bbf15aac5aeff1fd54cce"}
{"id": "train_037611", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "cbd612e6c6caf25940fb3baa90b9f2af24955d35"}
{"id": "train_037612", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "82567ede3948a6fad57a512f34e619d5b15e6211"}
{"id": "train_037613", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a6d8065b732f2f3d8a42d290c84149eb04346e2b"}
{"id": "train_037614", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "eb5a23b8215206cc4f225e6f7ca55e11c15eda69"}
{"id": "train_037615", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "5876c27b8b8384fd7539870f955c9c67a5c6f2c1"}
{"id": "train_037616", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "ebbcc6a5102de01d8256f16b19352a4f2736d9f8"}
{"id": "train_037617", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "32c3284233f56c00b69b27677c41d8eaf6b0b0ab"}
{"id": "train_037618", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "7c774d1c177c385c6357a123c78c7f02b9d4bb88"}
{"id": "train_037619", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3cc49d5e05d7418d2f3dc14c5132f3414a9af47b"}
{"id": "train_037620", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "745d9654cbabdb6a0455c0bf9c88476458f8998b"}
{"id": "train_037621", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b3037adbff9d621ae5bfec114bfafd4444f212ef"}
{"id": "train_037622", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "325ff28476c75532a739b6a716fad7c78d5945ed"}
{"id": "train_037623", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "99da31b3956b88bc38f3100da5f6e71d75ba546e"}
{"id": "train_037624", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "8638447bc76744dff7b6957a7e56976f4d7fb9c0"}
{"id": "train_037625", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "15db86b698b1fe4ac721d69ece33a7dcb5cf1eb2"}
{"id": "train_037626", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "56c527bc170411ef7b0c30f020a2b23857ed2870"}
{"id": "train_037627", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "df426a9905920dbf9c3b265b80a240bc9d35859b"}
{"id": "train_037628", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "426aea6f117d5ce95b830ba99c54e3769d6b0c87"}
{"id": "train_037629", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "cc2a006289a115c40dd8acbc9a77a19b0c8ac63f"}
{"id": "train_037630", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "f50e7693791c94c13ce2d0f5c9f37ecea4a63b0c"}
{"id": "train_037631", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d6b2d5cdcfd44c0adcded8916c59389254b4f409"}
{"id": "train_037632", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "dfa63af78b65c3868b08ae75f8c7fffa74be5da7"}
{"id": "train_037633", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "9eb7a802c41451eec5f5518b1411334c4a3fd2fc"}
{"id": "train_037634", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "14a754e74908f1ac461bf5292ae19030013a908e"}
{"id": "train_037635", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "fb3450052922af191fc49990d7559b2d103c5b46"}
{"id": "train_037636", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "da187dc434fad34d99ea81122701cbaeabb2e175"}
{"id": "train_037637", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b9948c99e40bcbdb8b3eea8a51858ec52b7f923f"}
{"id": "train_037638", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b386d3893531e2f9d1c36bb3b7a008ab9d88b33e"}
{"id": "train_037639", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "bf4117d0f42a525689c1ab5208720ce255d6a4ee"}
{"id": "train_037640", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f803fe0fe0e77d23c6134ad961ed7ec027e3a9b9"}
{"id": "train_037641", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "e8257ea901f453380faa8de3c963722a1b477822"}
{"id": "train_037642", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "503930869a0646061c735e796151fcad1847e2a1"}
{"id": "train_037643", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0247b493cdaf649323b959428413189fc9f36f2f"}
{"id": "train_037644", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "22290eb88faef093dacd284d70570960e8b93674"}
{"id": "train_037645", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ff483c90c3528ec331ad55aa05d377b06546edd4"}
{"id": "train_037646", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "39fc14bb3d94d09d053bef941f346da6bf5f5111"}
{"id": "train_037647", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "71a8faa1781fc6fe12ccba584169dca75d5eb286"}
{"id": "train_037648", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "769963f865948bedce8ab54d91f592cb71132b77"}
{"id": "train_037649", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d34566264bd00bc4f99d9b559fbd311c33256892"}
{"id": "train_037650", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "0aa009ef7e95813aefc03ba07a4121ed1bdf261a"}
{"id": "train_037651", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "9cb8ae02407f09b6f1af8b7929ad0f1da5e12809"}
{"id": "train_037652", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "150ee5eae867a71bd1231d6e6af7e34b87d598ef"}
{"id": "train_037653", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "70feea2450d77b079b359de2b8dfc74c433e89b9"}
{"id": "train_037654", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "af4b863b108751cb1f69f70c2c20bb7311f12436"}
{"id": "train_037655", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d69dd21840c50244ac55ab9f1aac4dd3092328aa"}
{"id": "train_037656", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c343390438a6beab0890526f1602f35e96ca46b6"}
{"id": "train_037657", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "11500a44c573016cdedf9cdbdc2e4df5036d28ec"}
{"id": "train_037658", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f0e223698f55a3d9c8d29a0c6d98d6ad52371f0a"}
{"id": "train_037659", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e84d279fa854069d11ee228c616a943c19f400db"}
{"id": "train_037660", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "8cdec2ec28c28b7b567b13eec0a4fdd8312d6de2"}
{"id": "train_037661", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "ff5ab87aea3c26ab63363d5e59c6cedf19d3bc0c"}
{"id": "train_037662", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "321b2e5cdd2cfd47828bd70c1684f342871b049f"}
{"id": "train_037663", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "1e5107ad8e466c8ec1d423453460e8e3fde2b84c"}
{"id": "train_037664", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "1edbeb384647270afd0e8fb392cc5e6d76d35857"}
{"id": "train_037665", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "9776424adb86bfdb54570d94c7d71063cf1fe845"}
{"id": "train_037666", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "f4df679c7ba202de10ae072a05117b8d2f57cb8c"}
{"id": "train_037667", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0ce255d001177f89df2d839945b60eb6c596cf5c"}
{"id": "train_037668", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "47c44788dfe583927f292581df4759f329729a89"}
{"id": "train_037669", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "8baa0bb7546d01bfb36706e14f79551294f63680"}
{"id": "train_037670", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "1e43a9eff46635a2597ac611010e182e591d52cf"}
{"id": "train_037671", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3227127f907d6c86a6567ec4902bd13ca9b4c4ad"}
{"id": "train_037672", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f2bb02baf63497ccf1a715caf6e714d78f05b6a8"}
{"id": "train_037673", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "8e6659dcd448120f059da1486c74635b38c36fb8"}
{"id": "train_037674", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "8d5523318ca3b8f4e6a3d04bd627045720185fb8"}
{"id": "train_037675", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "7dc51a48bd87ae4dcedd275851bfa2065e753df2"}
{"id": "train_037676", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "44b604d04c085eeea2e6adf09d68c82b458035d8"}
{"id": "train_037677", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a613372c9ee58cc07d7afd55b496357d8b2a437e"}
{"id": "train_037678", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "759aa86a974fa2a7760536581c1ac053fe603dce"}
{"id": "train_037679", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "1ab0a890876a7c3193dae367fea1a08a6c5f7629"}
{"id": "train_037680", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4d2f523910eae2666131b1b39e766cc003cec987"}
{"id": "train_037681", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "92ec9e0323b13ab4bc5dd9a629a389122327250d"}
{"id": "train_037682", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "9dd5b0b6af59146721ac358edee1676e04ce6189"}
{"id": "train_037683", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "5370328fe2e580dde8740fe250361a0158f976ca"}
{"id": "train_037684", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c3c6a4884c687d12d40690c0a9ae6bad8fc0d514"}
{"id": "train_037685", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "2cefa1f301cee12d3084a0050f3fd8b968a435fb"}
{"id": "train_037686", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "5200801a6b22b8a4ba7813b1b65374fd229886d2"}
{"id": "train_037687", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "914deb730b1ff5a467cc12bdb81612c5e3f8ceda"}
{"id": "train_037688", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "dd035b1316333fbbf9fa41aa9e96b5803eae4d6e"}
{"id": "train_037689", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d648a03234bd8468ab72e3d5ec932f07a8887e56"}
{"id": "train_037690", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c930d1e8383e7b4f1d51501e223fafd88c7521b3"}
{"id": "train_037691", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6615c83843b2259fb2c9be3b97b91deeb3991cc8"}
{"id": "train_037692", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "697eaf47cea15e6783a980363e82208c91db4969"}
{"id": "train_037693", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c26fce838262a39c6fd4da4504f0efbf76db53be"}
{"id": "train_037694", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "0cdd826af0f402689897c1dba1842aeec8bfb469"}
{"id": "train_037695", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "8885e08000ee1972f0678cbb702a58b859ed418f"}
{"id": "train_037696", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "b8e0c259ffa6dae9076806ed35691c3435fa2b7e"}
{"id": "train_037697", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "03a846664de842c11bc577767d061582b8682f93"}
{"id": "train_037698", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "0b0322fc8ccb919922117b56f26ba832afb4e8fc"}
{"id": "train_037699", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "561625e91056a02c3383ff8bbf9febc907ecae4f"}
{"id": "train_037700", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "98564065d328819074191e11af5106eb4b6a2136"}
{"id": "train_037701", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "6a0aec13635bcd75069d97640abaf66d564fd226"}
{"id": "train_037702", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a617c329e2695030284c6bc23492ed1ba84a5d88"}
{"id": "train_037703", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "2ff832f76be904ebe87a2e9c0df34018624ec313"}
{"id": "train_037704", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "130088ba904ee13c3d9a1bdd9e105d70b3d65a5e"}
{"id": "train_037705", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6553aeb8591f04841f8d5cdeaee3487184c817fb"}
{"id": "train_037706", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "11daa22d8d82eb9df3fe1700ef19b9a69f993052"}
{"id": "train_037707", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c5d0b6a72428fa21a4b004962e6507df6c5d34db"}
{"id": "train_037708", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "036267d242927af0b55efc8c9e0a69297af42b27"}
{"id": "train_037709", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "4f1817cc1be69b7662fdf1de4da125570568dd80"}
{"id": "train_037710", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "200c4a404e4bd3681ee7f8c854392f443e8abd3a"}
{"id": "train_037711", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "fd1cc7dca40914c7c33140ba3552172d556d5d68"}
{"id": "train_037712", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "eafa52dd1479dd73d483fcf474393b257cc26dc7"}
{"id": "train_037713", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ec4791bf1adfdea36d4d9557489676082387e6d4"}
{"id": "train_037714", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "489335b0806d7dadfafe9ff7dc3eba0302c8f09e"}
{"id": "train_037715", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "79021d5c633e23a03c90592a06eefde5f0f949ac"}
{"id": "train_037716", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "d8847f03c0800f45b6b7d5eaa315540a131905a8"}
{"id": "train_037717", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e3416a1616d16d49ee337a45c642004fc4ec9f8f"}
{"id": "train_037718", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "dfaab52e21e66783106b486b099e432eb9142b1b"}
{"id": "train_037719", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "8655c7f381305370f444320e5a92ec83a4ace1ca"}
{"id": "train_037720", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "0146d43e27a02c79f6669d881a73462fe84072f3"}
{"id": "train_037721", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "50d0e2a0a4fca380203099e82c04651d203a66ca"}
{"id": "train_037722", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "81ab849d1e608a509c358a854a8c7aa78c8bdd18"}
{"id": "train_037723", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "86b093a2227bb72b707369eb98ae4f71d307228b"}
{"id": "train_037724", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "5e46b4e82ab72279bc625cf5d5a4c8e31c29a081"}
{"id": "train_037725", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "224f43f3dd065b285f1980b51e2fe9949e51eaf5"}
{"id": "train_037726", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c49e5e65d23297c65474ad623076fa22104d57dd"}
{"id": "train_037727", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "d238878ab9a7719d2479281c3aa621f2da93419e"}
{"id": "train_037728", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5c19c298a40521332a0a7a1e812cf9fb5faded80"}
{"id": "train_037729", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "69259d1b4c6045afebb2e8eb141668244200bf6a"}
{"id": "train_037730", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "9479662d72f23b9bbd92651e4825490eef11574a"}
{"id": "train_037731", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c970982f979f3086122a23d78cd6703a7cdd71d4"}
{"id": "train_037732", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "a3520f0389e7b5c990ff2df35c80c91e0a95141d"}
{"id": "train_037733", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0ce255d001177f89df2d839945b60eb6c596cf5c"}
{"id": "train_037734", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "966e1d6fa34d85ec62f3e39b50978eb426034b01"}
{"id": "train_037735", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "a42521e1acd4fbe6326682a3541d49e609a433d0"}
{"id": "train_037736", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e8521ce42704a85cc4efda8ecd66e3060c9640d1"}
{"id": "train_037737", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "cbeb7691dd0726feb4bf549ebde21a1b395dfb03"}
{"id": "train_037738", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "da0cd82555e9c25bb076b84a509a5adc83095151"}
{"id": "train_037739", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c08deab629be508a1a0388feadc90bfbd3274aae"}
{"id": "train_037740", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5e526437c3773e7d02cb829d299940f82948bd64"}
{"id": "train_037741", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "07c0344e5297cc64e25cacd3f938345aec70b671"}
{"id": "train_037742", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b037c293bcba0ae3ed5c23756112e2702c460ef6"}
{"id": "train_037743", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "d8521576ec4cdc6922c888f12456a5487720930d"}
{"id": "train_037744", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "1ad37656e4e7280754148d8bf8a4ecd7a655cf5d"}
{"id": "train_037745", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["agent", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c8a52fd979e5766c72579d1cd184a54d52b4a16b"}
{"id": "train_037746", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a42156249d6664854d55d930e14d03b9f06e912f"}
{"id": "train_037747", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "039684440b48c9e76933916995613f38b4d66501"}
{"id": "train_037748", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "1974b0beb7d5c2b7c84887679406e85149fab38c"}
{"id": "train_037749", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "1a3ed2b9f8c71abefea20eecbec7c1e8dbfe85db"}
{"id": "train_037750", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d67ae2ae2f4192822eb8d0e9682f292a9e2ec036"}
{"id": "train_037751", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "09d65becb6458c4ca001198e0492295b0fa3b805"}
{"id": "train_037752", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "1e6405bca8aadcf0d23bfab7f6a5567604004e8a"}
{"id": "train_037753", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "daa601f7aee3258bca0ccd531ec1abb9a63f8a37"}
{"id": "train_037754", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "1599889f21310b4fe8434041a4babf9a72238395"}
{"id": "train_037755", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d5c8870c80f696e1bf399c4eb32ee63df7e403f8"}
{"id": "train_037756", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "300521913debacb33ab6fe66b11ab04b64589d3d"}
{"id": "train_037757", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "0b89b51f7ca64b0698846c65e0c7fa6d268b075e"}
{"id": "train_037758", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "71492e4198a87e4931939add80f00163b79740ad"}
{"id": "train_037759", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d376144f499d80b63b5db57aaabda1b8c91954d4"}
{"id": "train_037760", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "ce2126bfd7d97c5b4ce02326a104396ce86a9936"}
{"id": "train_037761", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "4d630552eb4eb359de5dcc7f8ae1bd4bfa353a23"}
{"id": "train_037762", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "3b1c419eefc5e67507c4cd05a8f54090cabb49de"}
{"id": "train_037763", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "6e43e4e8f086f211a2530e519bbff016075a4370"}
{"id": "train_037764", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "d17a1af3ce29ffacee977c920568b9b00940b07a"}
{"id": "train_037765", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "072d64ce27be2199514d04771c5a1caeebdb22ef"}
{"id": "train_037766", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "a9a3d85a39f5ade41a66d7b082d1d29f7c3ff94a"}
{"id": "train_037767", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "e785e16998e7f344b33400c0d953940b68d68086"}
{"id": "train_037768", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "e9d3eeeb9549715841ff508c31adc4528d2005df"}
{"id": "train_037769", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "46e0e54e693bf39137d83468cb28c8dacaaf23f4"}
{"id": "train_037770", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "3224125a421679abb5664020ef6f76d1e150ae5e"}
{"id": "train_037771", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "52bc9ecbd01a8c04c67ace07a85d7b496436c810"}
{"id": "train_037772", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "dc4bdfd6e750abde08ce8cb481feea5178a5992d"}
{"id": "train_037773", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "81aa9df93fe3dffdce808d25b5db0c712ee9d5c1"}
{"id": "train_037774", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "076f414cd5eaeeb480d35c719e90c50a5c1e0392"}
{"id": "train_037775", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "96801fd64cdf90b9beca79cdd96085dbb7233ca1"}
{"id": "train_037776", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ae75c0dd65fa08695e54ed88f609e0fa3e4b60e1"}
{"id": "train_037777", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "9851f7aba65dcdd9117d7edfe7bb7fc30827f8cb"}
{"id": "train_037778", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d369cd400b73ff6622e619832f758f86cb470bdf"}
{"id": "train_037779", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6b6440a740ac6f42979654eb66a7d22b3711b853"}
{"id": "train_037780", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d4224f7210d9467d5fcfb806a7c307ff929bcd7c"}
{"id": "train_037781", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "5b68c16d24858fa5d4a95aaa85926fc0d363de6d"}
{"id": "train_037782", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "031ace4e56aa79d17d7e30a4a2371cef1c58f2a8"}
{"id": "train_037783", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0d4f6454c7750c5ed46223531b0fe04f2408165d"}
{"id": "train_037784", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a11295f4563cf8ca9fc933785c3614521fa2ac74"}
{"id": "train_037785", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "70d78986822c7500b779b5d520adf99efda79cdb"}
{"id": "train_037786", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ab54060901754270eb4930d2f0d9ffc67130772c"}
{"id": "train_037787", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "06966bd304fa1a508b00dc89b39a358242d2810b"}
{"id": "train_037788", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "9f2b84ecc0e1c66fbc94dbc1c17f3f8e3d5bc4d9"}
{"id": "train_037789", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e44878f6c76ce75a38b91b6f9a21a9def6c63bdc"}
{"id": "train_037790", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "9d902bad9a6859aff4134e50812098af63c14500"}
{"id": "train_037791", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "b5b97af04c6b2d33bfd95217f16f50e8adb5be86"}
{"id": "train_037792", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "46e99d60a6d20bd254819bb3146f43cfe6e02962"}
{"id": "train_037793", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7a06db6fa5b6fb5462b49fd1516e9019d7b6046f"}
{"id": "train_037794", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7e9ac79c468ae04ea5c695e2945da60cf82df0de"}
{"id": "train_037795", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "67a140477c356c9318053545ed901b8e0e4a86e3"}
{"id": "train_037796", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c2e17d1eaaa67204f398d73e91dfa95b526b62d8"}
{"id": "train_037797", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "1dc17d23863f84c557acb068df52926e8deb27a8"}
{"id": "train_037798", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "1d82297ee28c9c269ea502c7013fca1017f852e8"}
{"id": "train_037799", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "4928e81846125193202cf0840293bf0c58ee82ea"}
{"id": "train_037800", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "29baf611f0dfafbc76fcce2be8b53ff5d2ac3203"}
{"id": "train_037801", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "41689ea55394326a6ef7964d38d8c29fdc26bf8f"}
{"id": "train_037802", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "3842e794f6203812d0ff519da089c755033ffbbd"}
{"id": "train_037803", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "290d97891e1e978141e29ba7731cd4985665b448"}
{"id": "train_037804", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "56c774ac28bba89ae1f6f4aeda6dd255b5219a09"}
{"id": "train_037805", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "e04243f825474e5a3792a36da3a600ab0bc67c70"}
{"id": "train_037806", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "3332cb2f6d470424f2dda572b42edb68b7f2f983"}
{"id": "train_037807", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "a0e8a75be50dc8179d09225566c4733e782b5634"}
{"id": "train_037808", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "010f6d3c9bde20e3cf626d1a8efdd43dc1076c0d"}
{"id": "train_037809", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "9954005050a413d512faa17e3837a5f49beadabf"}
{"id": "train_037810", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "fe2bd2e1ba29d6e619955de07be50574c5f5962f"}
{"id": "train_037811", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "5bbb3495afda73c06c17adafb20b90f17cd3b7e2"}
{"id": "train_037812", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "6cdd45798d754268adc4cf49d5903b266fe815f4"}
{"id": "train_037813", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "02172eb5e6ee2cc5ae142eff462683a5b4767eb7"}
{"id": "train_037814", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e836fc1c13db94105cf1ad1f87e16b8c52681cb6"}
{"id": "train_037815", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "ecf68be4913f3de8b49f6add502a2e3dc4802d64"}
{"id": "train_037816", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a7e5c6a9e8256bc992c87344376bc46f12140dc9"}
{"id": "train_037817", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "50f7db09183e823c2c0d7bd9f8b7e43899770d0e"}
{"id": "train_037818", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "b9150124c9f85120997d2ba649ca7bce904ca03b"}
{"id": "train_037819", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "725404e627728ace187e4e49295eb7f4073556a3"}
{"id": "train_037820", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7d036c28559e6091b60edff3a61c2156014572c9"}
{"id": "train_037821", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a373e65b80acb695a12ae8427e568bd3c7280f3d"}
{"id": "train_037822", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6292c8c8630e076e7c81691053f455c35281f379"}
{"id": "train_037823", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f4da8445aa03e58d11c569b3754eabf174c4a399"}
{"id": "train_037824", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "d225b742e01776cd9e97b49b4a51f42e0061f357"}
{"id": "train_037825", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ac98339a542c716d97f1670d5b66839446aad92f"}
{"id": "train_037826", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "db8d27bd40f8aff2fa541d298e7a4c727a7e249b"}
{"id": "train_037827", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "231a9933ea30bf14c13dd20157a736730e9424bc"}
{"id": "train_037828", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8a6bdef1422a98cac26e4a84bb5e3f9a232e202b"}
{"id": "train_037829", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "969d832e43a71ec58e885144ff560db45ee5fafb"}
{"id": "train_037830", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "cc25394b54a2b484194354f497c0d285c4fac3b2"}
{"id": "train_037831", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "2e86bdc7fa3d95ac2cd676f4e16bee7909138b94"}
{"id": "train_037832", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "ec8ae4355e38d39b82e6ca03358199771b84d488"}
{"id": "train_037833", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "9b6cd909d7c19867d12c26066ef5de3763d86666"}
{"id": "train_037834", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "7a3115815b4e4bd72f2da3188df116b00c0f3503"}
{"id": "train_037835", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7bbaea093756aa55e59dfc0a2c933c357236efa5"}
{"id": "train_037836", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6a6fab19ac55154ccacba55832fa1d919c01faf6"}
{"id": "train_037837", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f97ea717caa30943a7bded196b55b56f9336c82c"}
{"id": "train_037838", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "5aa9ab865ba54b52c68de2b63975146e24700a6d"}
{"id": "train_037839", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "1feda04145ea2a14cd431293f6b8f333c6a89b78"}
{"id": "train_037840", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "2dd44c2fa6dccceb79bb24fa86cf5ab2d7027029"}
{"id": "train_037841", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "eab556d42943bed383a13b09781870200bfd5301"}
{"id": "train_037842", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "9be64a00c73c98aed0dfbf9ced46bd37c98bc9ae"}
{"id": "train_037843", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "5c774e01dcf306636e19e7abf36626aa008ca94c"}
{"id": "train_037844", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ddd759fcd5207081ffc4f05798c17b54ddfd5e0d"}
{"id": "train_037845", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["governance", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "5ae4d946b284ef00b32237a636fa751c5f19cb17"}
{"id": "train_037846", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "943fc3e3ada7577de313bef03f5ea82caafe8fdc"}
{"id": "train_037847", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a0770877c2334a374e08c7830627700460e0d25d"}
{"id": "train_037848", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d2538b0b141d28d70bce2226589742393d1f5bbd"}
{"id": "train_037849", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["research", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "412e883cf7df0cfe5d5491fcf90827868c189672"}
{"id": "train_037850", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "871292e406ca2d3bcdbfd24722ef592f1513c515"}
{"id": "train_037851", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "1f2273108d9327ad9e8778499255bb50cdb316f9"}
{"id": "train_037852", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "471ed447bc97b72f494ff2eec2c880948ce38f81"}
{"id": "train_037853", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e2a2e0eabbf6f23a3ee2d778e805a06808401171"}
{"id": "train_037854", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "234058a11fd0fdb1a307bc8c0f3995c8fa775b91"}
{"id": "train_037855", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "65f41958e0e8b016bfc1d914ea4612b90b662270"}
{"id": "train_037856", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c1969ba3826379e5a9465077d854fd490e3a68dd"}
{"id": "train_037857", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a537ee2978e640c477bd26bee97abc671729c521"}
{"id": "train_037858", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "6502d79da12be17c773583b437f8daf424e3aa55"}
{"id": "train_037859", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "9447561f4d5072f07f786335afc9d8c8ebaa71e5"}
{"id": "train_037860", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "7879a1ceaca7d6799e49ce3ae7fa90caf0e49148"}
{"id": "train_037861", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "4d92699185a0cd76fc342b4600fc1ca7f6d2c79d"}
{"id": "train_037862", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c0a73a0a2a665546213e5734aa481e15f75ad732"}
{"id": "train_037863", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "315a17fca7effab7154400a3725d74dd90cc6a1d"}
{"id": "train_037864", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "804f8fcd9a45ec30c54134bfcd6bb0fdf19ba97f"}
{"id": "train_037865", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "00076d9e9afdc29e93236443486b4d8ab4424c18"}
{"id": "train_037866", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "e2429d6a2dbe7b19463c6d14404c0e75a1c08376"}
{"id": "train_037867", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "bd1bdc736389d468077cda22988808b0894dabed"}
{"id": "train_037868", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "b4f8ae07aaac32f086c4256be8d78337e96748fc"}
{"id": "train_037869", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "354149dca4b1b5ce90829ac0f5bd7743e30a4f88"}
{"id": "train_037870", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6420ab532a94800a133984c3c0b929c0b3d268cc"}
{"id": "train_037871", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "35185c95d8c5bc0ac37645b56c46c281b2b9af4e"}
{"id": "train_037872", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "8b9a7ed9440ae98dc5dabeaf6c2488d3b51b7376"}
{"id": "train_037873", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "65d666c730e3920275dacb5eed4aee35428ff585"}
{"id": "train_037874", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a26d367c085e0c10a7cfe4dcf015282567b1ee21"}
{"id": "train_037875", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f29fe2eeafde04a37f274cd193bc15392aec420f"}
{"id": "train_037876", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "7ea14f6ee0567321f547bf0813c3ccad7349c8d2"}
{"id": "train_037877", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "452c2dd8a150633dfd6cb42e17df3ef34b80899c"}
{"id": "train_037878", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "b3708c8064e13a002c35dd6ceaae29b7fb4854b4"}
{"id": "train_037879", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "4abdbd5eaed653dde60ccd8a5de1bce2a188f706"}
{"id": "train_037880", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "5a02f8e70cbb57a3dc37bdd435625f4440b42859"}
{"id": "train_037881", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c9c58773a3d8c8b7ec9a52db05e33f1a84bd0813"}
{"id": "train_037882", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f16d9644b020f03faa36cea6a7363a7cd50df594"}
{"id": "train_037883", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "2329ce0ba2c5a3f3e218dc80d24369a18e09fa76"}
{"id": "train_037884", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b239c01479edba64c17f32f13ee08a9aff7f28e0"}
{"id": "train_037885", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "4c7ff7a8dfd0787332f0480aacca9cec93b69d30"}
{"id": "train_037886", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "6ec0a6422410d5407d08f9b35580722cfd95379e"}
{"id": "train_037887", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "2e842c9adfd0ff4fb35e4d60d45b2d11294f8499"}
{"id": "train_037888", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "8fa7e2988699f1ba25d6cf73bac2c9bf214d3993"}
{"id": "train_037889", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "2c87e357350b91fe09c0bb63379e0204f9712063"}
{"id": "train_037890", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "c4fb3d690efe3905da4cef5456725c9e91168811"}
{"id": "train_037891", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "1ad37656e4e7280754148d8bf8a4ecd7a655cf5d"}
{"id": "train_037892", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "1b97eb963f1e8ee0072b68a8d1d765721804590e"}
{"id": "train_037893", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "e512cb7d3bbca18c722a2e5435d06f2cd5a583ac"}
{"id": "train_037894", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "0759c1b247355b128cf80108f90c775700e968c9"}
{"id": "train_037895", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "a7330d894263bc0514489b0dae5e41c978157d06"}
{"id": "train_037896", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "5200801a6b22b8a4ba7813b1b65374fd229886d2"}
{"id": "train_037897", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f140de423572970733114dc96f3b78a2bcdf7820"}
{"id": "train_037898", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "23ef63d51d8b2fa6430ed478219740324af74202"}
{"id": "train_037899", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "942a43b8cfab0645b0220c6a5bb228f22555a5b7"}
{"id": "train_037900", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "562445e7206bee127bd2524724dabde0a1ec022c"}
{"id": "train_037901", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5594547fde1d5733d2fdd7136f5aeb23310bf61e"}
{"id": "train_037902", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "24f20415325ca18c2901a592b51250df02e13e37"}
{"id": "train_037903", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "d1d13a07ec3cb1dccc57be1026ffd44a98fff5f1"}
{"id": "train_037904", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "48c6e7056e8bb9c328219cada0b3fe293cda7c7e"}
{"id": "train_037905", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "0c01ffe135fc7fb7e93e26461cb4ab52552fc3c5"}
{"id": "train_037906", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "cd5f845724fefa014580b3bb9956c3486c08fbe8"}
{"id": "train_037907", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "dc2e1e9a71437c14b29274b6963e030578973f52"}
{"id": "train_037908", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2e34991ba666bd2edbb91351ae0831ba5e5921de"}
{"id": "train_037909", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "08fd6adb2530ece689e2e3b501aec81239ed10c8"}
{"id": "train_037910", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "348fe4e0e893334d269e3e4c9adf647b24814766"}
{"id": "train_037911", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "435bfe5ba13aa9877a01dc589ab46ace720b62ab"}
{"id": "train_037912", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "960dbc67c167704834089341c6b69802f7c806ea"}
{"id": "train_037913", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "69c624b1e45658cb04eb8801d613386d72be7683"}
{"id": "train_037914", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "15db86b698b1fe4ac721d69ece33a7dcb5cf1eb2"}
{"id": "train_037915", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e31a6e98fe6002f345cdc4002d7d0cdb51d947b3"}
{"id": "train_037916", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "33479a1755023414213fb0a660e5d7805640cb76"}
{"id": "train_037917", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "f44bba0cb3b8e735794ae260e84f9d73db20eba6"}
{"id": "train_037918", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "0d712c993ccd69fd493c7e3450dc18e1027f9071"}
{"id": "train_037919", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "76acfaf04666d8cdf5288bfd697d83ff325e614d"}
{"id": "train_037920", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b46e1258decae7e87b7d38b092a0691c3cc875e9"}
{"id": "train_037921", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4ba48dafde2ebecc5e3b004236ccd4da6ba3a622"}
{"id": "train_037922", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e0eb024413b6779bc73e3f37a03aa74298fe908c"}
{"id": "train_037923", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "8c09f1d6980098168f57c45b6b8418a633b5f204"}
{"id": "train_037924", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f2499a642c9ff7a52f0e238dfbfdf86167610209"}
{"id": "train_037925", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f08a816430e49bfe7f97a9fd10bf4bcd22f3ab83"}
{"id": "train_037926", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "80cb6d32a77f68d7ad039a2cd4b2c0da5a44e566"}
{"id": "train_037927", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a81fdb7cbb37d53cf816597781881405118d5b33"}
{"id": "train_037928", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "239c7d769aa0298e14ca98efa10c175ed3bec185"}
{"id": "train_037929", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "9440882cc2046da254895f86201e4e92e53ae418"}
{"id": "train_037930", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "d905ffec603820094dcab5392a362ab5bff097ac"}
{"id": "train_037931", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "3e382cade3e60b4a74deacf670065c6697854e15"}
{"id": "train_037932", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "850769d414199edf89e7bdd3170f7dfb12c22c34"}
{"id": "train_037933", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["coding", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "31e5b94748d226097b72f9be79114c43829dbb44"}
{"id": "train_037934", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d698d5a8fcba9f357c3c94e3b4e31113340592ad"}
{"id": "train_037935", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "428f962dba31e767c01b38837c9f85c08b8f04a0"}
{"id": "train_037936", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "8e58b4e3a3086e9e99a12d0bbb73b15897034c90"}
{"id": "train_037937", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a777e20a32f55b6841c451a58edd344d365c583a"}
{"id": "train_037938", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "486c4853c8bb7ac48756df7375dc6a57a75b5f5b"}
{"id": "train_037939", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "0b8038a058ba082e1e1f3eeb015b621ecbab0083"}
{"id": "train_037940", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "8850c06c953012657e66951eeaa94a6ddf5714e6"}
{"id": "train_037941", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a84b4694328591abf9342a67881e7a7bac73f686"}
{"id": "train_037942", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0ee23d9a15ebe5c17563768488a7629169d3bb9b"}
{"id": "train_037943", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "eaeca5c1d241349cec1fdafc0964bc577e35ba06"}
{"id": "train_037944", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "4ce183bf3e405faec737a44e9bcf976acbe45b7c"}
{"id": "train_037945", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "1103c4f6db7dead327eee53df6f9e232e0b17766"}
{"id": "train_037946", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e565d5793a4a21c459bfed9d353b035e7a3201e3"}
{"id": "train_037947", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "144379d37767c9b0c8b609782e359489f634b7ce"}
{"id": "train_037948", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6243243945e8b1e2692bfb6c268b36c2e72c79b8"}
{"id": "train_037949", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "519a6271730a0ca3edacf50b62a694cfb845e254"}
{"id": "train_037950", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4b75459e475b10d81c72a1ef73e06bfcbb42456d"}
{"id": "train_037951", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "53f7ecb79abcf4dc819c461a06c8a35b7404c564"}
{"id": "train_037952", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "3b617f782e19ab70abcd23d177726303846f1c8a"}
{"id": "train_037953", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "fa52ee6d443880d28d2f2ca14ea5b6036455dd9f"}
{"id": "train_037954", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c6524afa1c68875d5ee91d143bb2cc119d636694"}
{"id": "train_037955", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "57655d196b85ab47c0827617bdf66ed191ecd900"}
{"id": "train_037956", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "20e8766b94a34efac577dbfda7f8fd0aad80ebf9"}
{"id": "train_037957", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "b20522ccbcb561ec8a43639ca41963fe8e8b2508"}
{"id": "train_037958", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "7dcc407b97997a41df05e0c4747720e7fae01104"}
{"id": "train_037959", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "ffe7c5c4eafcea71c10f36e7075cc0bf4f1684b7"}
{"id": "train_037960", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "6f942639479c07786c5063cca45759dce8881505"}
{"id": "train_037961", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ca20ae9c072925546008ef1ab7c5b8a2a28f4508"}
{"id": "train_037962", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5d87870bf356d9df889bd5100af2c818ea719a73"}
{"id": "train_037963", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6e833dd627bea24432d3a3424a7003456a4dafb2"}
{"id": "train_037964", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "c1f42b56675e7a3fcf2de25924652fd6fe0ebd6f"}
{"id": "train_037965", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "4f0d1b7ef9f6ab81dc390b7b88366d8d3367a269"}
{"id": "train_037966", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "26328bec27c00ef77505b872b1051554f5c16c70"}
{"id": "train_037967", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d71deaaa62d71b4144410ea580e5cd5c1c61b90b"}
{"id": "train_037968", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "b09415c9ebf6e5e3f25a7f8e82109d49370e16e0"}
{"id": "train_037969", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "85fc53b09d7af8c09567edc87b40621a9f087722"}
{"id": "train_037970", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "fb640ea49a04c5347f656882df8d97e583614e8f"}
{"id": "train_037971", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b93812e2c50be8c3605cb16542358acbef6e7394"}
{"id": "train_037972", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "431e307ad096f4ff3ff06fa9a056b49798f4e9fb"}
{"id": "train_037973", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "370d407439060fd327994afbfcbe6b77bca140ab"}
{"id": "train_037974", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "cb2bfb4f47fedd23c286028f98ce8601e03ef555"}
{"id": "train_037975", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6862d7f102f1225beaa089aca9087bbcbdb9756c"}
{"id": "train_037976", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "730ebddead443379bd25b0b721b6c866e15e4acc"}
{"id": "train_037977", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "5e02800af71c0b4d014b3647ab5bc691c4938229"}
{"id": "train_037978", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "bb37b2665fde83547b489f413ad52575a453aa7a"}
{"id": "train_037979", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "db419701eb9b8819898d52379cbab52e782d18ab"}
{"id": "train_037980", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3d39381c7e38efdcd3fff8be7440da05830e10bb"}
{"id": "train_037981", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "35367794c3f605f9ec62863892ac9f7c7d099367"}
{"id": "train_037982", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "18cc83299418af2df6a6ab265a8664b270bbcfe1"}
{"id": "train_037983", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "4d2e56c3f7e6bcef4f8eccd9a4aa7eb337561fcb"}
{"id": "train_037984", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "13a6b5c6861160c4b5860b35a0d0c8d49f8cbebc"}
{"id": "train_037985", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "1c7388c4977c3c72d07279c2e864eb410ee31318"}
{"id": "train_037986", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "28cf1957564fe753dd753cde10125d5249f4e471"}
{"id": "train_037987", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "20a0f4600d31969dcdfc1a09b81a31a8c09f8c1a"}
{"id": "train_037988", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ce3f80022d47ab1023b3d7e005c069de44850206"}
{"id": "train_037989", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "36d1afb3beb8d5a5f273fdb10be6d38898b1143d"}
{"id": "train_037990", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "2b63030313c76975c57428b71ea235c9d5910da2"}
{"id": "train_037991", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "7f488ba2ad69da98c8b6fffb7f32e4590509f1ca"}
{"id": "train_037992", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "2c40bf61ee35834cc8c94a399da7390d2b8cc904"}
{"id": "train_037993", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "6b0d4829439a7303d2a1d0756386388c656dd16c"}
{"id": "train_037994", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "5ba95209d69adbafc8a074f3b9033eface247980"}
{"id": "train_037995", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0c1f06af552e67875c73e60bfd4d7fdf6ca0be4a"}
{"id": "train_037996", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "9d0ac12eed1779e9eae63151c1e72c02470642b0"}
{"id": "train_037997", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "199b1117b4ffc10cbb51ffc25614fb0713664c6a"}
{"id": "train_037998", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d9019e54144d513edf30a3668c969d8a0742d20e"}
{"id": "train_037999", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "33a032b1c2abf5c6527266460cb36ed66b3fbaaf"}
{"id": "train_038000", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "2313d27830bfcdecd3f855be90bff03fa62955d8"}
{"id": "train_038001", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "07a2af5e8b18a939d7ae2b54485017afbc65257d"}
{"id": "train_038002", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0e609322e07e53098effa126d38934300b0a06e8"}
{"id": "train_038003", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "81ac9a92638657745501e6ed7a8c89fd017b528b"}
{"id": "train_038004", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "8165940eebc3729b93f5affc2a88393e91641712"}
{"id": "train_038005", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "456f7ec74579d7eba33b10548bd3cc358f249c85"}
{"id": "train_038006", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "0ba07aa5c86f00bfeb9e2e74517fd07354a067e3"}
{"id": "train_038007", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "5eb0b57e12c1c1e2ce2f5c3375f98ebd13d98abe"}
{"id": "train_038008", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "44a8afe560771e710e3c5320f279bc1909af61a5"}
{"id": "train_038009", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "65db213be565a4692ce940dc6db82dc431d1734c"}
{"id": "train_038010", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "e001a2f4c9446195168209942bb0a87e8a4fc39d"}
{"id": "train_038011", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "8d20a7497aaa2edb4a4206ef083581f739f1d155"}
{"id": "train_038012", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "67c4d9ef419aa976a6ef53fb2a786df6a960f2c8"}
{"id": "train_038013", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "195beffcd97e1b57aa813b2c5766ccec0abbf522"}
{"id": "train_038014", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "cd0da55f720a19937ce71a79893878a6ba376b5f"}
{"id": "train_038015", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "2df2ea66ccd88419311464b45352a43d79323bdd"}
{"id": "train_038016", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "377d8df923cec92532b8ff8a0335768ab4c135e5"}
{"id": "train_038017", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "48da2302a2b24d2e8f827af8612456dbf9ba5564"}
{"id": "train_038018", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "852c0c1adc0bda9811a56f045ed7d40dcc0171be"}
{"id": "train_038019", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b88ad6ae525d1721e46b76230c1c087675af08e5"}
{"id": "train_038020", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "44005b060bb8add469e47563fd57f9a315656a65"}
{"id": "train_038021", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "30c63915e2a431a60049b78cee673e5603b1c33b"}
{"id": "train_038022", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "300f9d1bc5865fd3561d0f513a306c925bdd27e9"}
{"id": "train_038023", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "12d72501c3b2948ac2d225a2f704fcd639186f7d"}
{"id": "train_038024", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "0073d9f517528c565ec725ac49f21a6e11dd687d"}
{"id": "train_038025", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "eaf09b41539299b2db2d8353f22670b6fd5af487"}
{"id": "train_038026", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "8dd6650f3003bfc36d31e4081042889483f52468"}
{"id": "train_038027", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a4904a44b1a1cc1f1c4c3f2c7cd3ed6c6bee9bc8"}
{"id": "train_038028", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "02561af755791af61118326abc59ef9c13e56643"}
{"id": "train_038029", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "fb0db58d102eaae5c37069af86f67dfe498e2e71"}
{"id": "train_038030", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "e36512f17f8586d789576dfe1d86eccc38e59da4"}
{"id": "train_038031", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "dc68260f97746fc7b4697be1b5442ec10be85176"}
{"id": "train_038032", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "13b580eb16f6234c05cb186d2aa2f51ea07e5d6f"}
{"id": "train_038033", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "0bd479ec7248877fe5cfed84f2a4a43387b24d96"}
{"id": "train_038034", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "160463cee8d58546990e1076072b806b3d14e76b"}
{"id": "train_038035", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "8fb3aae6fa2ea3cb52d3d8825a4584dbc3f2bcea"}
{"id": "train_038036", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "80cb6d32a77f68d7ad039a2cd4b2c0da5a44e566"}
{"id": "train_038037", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "89709c1afb1d85db53f088dd5361634adf11f42f"}
{"id": "train_038038", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["research", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "8fb0a33b3879fcc4d65c562a5f7a8890e85e7092"}
{"id": "train_038039", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "61faf614c8564411fa0d117428efbfcce2f37b1f"}
{"id": "train_038040", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d42e103fd606a8408d4730580023dbc18080cdbe"}
{"id": "train_038041", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "2c40bf61ee35834cc8c94a399da7390d2b8cc904"}
{"id": "train_038042", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "2ad406382a090ae31a6591f49240b0ea2616f604"}
{"id": "train_038043", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "aa709c61609590a9cc349cd02bbd72f2908cfbfd"}
{"id": "train_038044", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "1418c71953c6dd0cb8550c4c86fc4ac99835f0e1"}
{"id": "train_038045", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "b8d4f02bc144cce05078098aaaf4d456f9eb0784"}
{"id": "train_038046", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b914a6fc0b8bc589011e30bdca7749da727182c3"}
{"id": "train_038047", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "27ed7080ae70aabe878a08d6f960df7fd21d77fb"}
{"id": "train_038048", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "1b66c4e0fff2f015d4e0158659712be4b7152acd"}
{"id": "train_038049", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "1b1390ac273f03a252e8d2700a0cf2dab7c6671c"}
{"id": "train_038050", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e8dfd42424d11d9e324683a2e04efc6384811bcb"}
{"id": "train_038051", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "45bb948b381a655dabef03521f4e689d1f04ab12"}
{"id": "train_038052", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "65b4ad841c594f6adb032d8ea5bcf079ec325c6c"}
{"id": "train_038053", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "41da8c97baffc49d4bc7032e1a8178110a586475"}
{"id": "train_038054", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ce6caf0ad86798d0e58db653d1f4e21d96801f90"}
{"id": "train_038055", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "fc60ed67bc183c310c64d794e4d2379fec69cdb4"}
{"id": "train_038056", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "54529c2f6ba2055c3020e03b62f2ba9b17c46b61"}
{"id": "train_038057", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "14003e8d5e9eb98ebe68885a4ffc1b299df9d0fa"}
{"id": "train_038058", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "8dea89501e07ee3a81fcf324dd03ceb70b526dcd"}
{"id": "train_038059", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "5fbcfd5d4ef79cfbbd52c05e9fe716920acf2392"}
{"id": "train_038060", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "539337e37b40c567588b5fbee2537481d57093e2"}
{"id": "train_038061", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "448b29b035826ea187eabcb57b16751ff11f94f5"}
{"id": "train_038062", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "171579cfce59abe5786fbd7e983226ddf647a9c6"}
{"id": "train_038063", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "23e91dd97ca412201230ece69adf651c83c29682"}
{"id": "train_038064", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "20bae42e9ff16d92cb1c1c5baeafb1a9c866f309"}
{"id": "train_038065", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "c8123279a6f5ce0c99d582409d6286995b34762c"}
{"id": "train_038066", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "4178464ee52a77f5d904e7bd478bde6f20f5f649"}
{"id": "train_038067", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "bc123bc4d9b22d24065812953ae25b8053a88229"}
{"id": "train_038068", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "569a48afbf4fe97430035faf4959b36b40ba6dfd"}
{"id": "train_038069", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "631d5965abb8d840258fc5b03b07bd3558eb2f01"}
{"id": "train_038070", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "cc6e3f41dd8b5440e60d965a7111de362666beb4"}
{"id": "train_038071", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "501fdb9574ced461ae8afddc7df576a1abaf981c"}
{"id": "train_038072", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "6d48f9391d5823d521b5d254cb72b0f2e55b884d"}
{"id": "train_038073", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "253b5856af155301401f711904bdca6f39b8afe7"}
{"id": "train_038074", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d8b858fcade573d9bbb63a963dbfc78b86427b52"}
{"id": "train_038075", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "05989c31f40a7601c220fc55940ba073b684ec9c"}
{"id": "train_038076", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "ec4a3320cc836c3705f41075d4f60b0b8a15ce3e"}
{"id": "train_038077", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "c9979df3daf8d79efa475341e79401469fb7b699"}
{"id": "train_038078", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "4b75d31c3d22775b6e33d685e953ebc63bc1fd9b"}
{"id": "train_038079", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c4023bda6e0b9090701791164e600733e8f434be"}
{"id": "train_038080", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "5a933a8831506f9f7e8aa756b369cde57e368915"}
{"id": "train_038081", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "277189aeea628e93abb491910e2020c1812f4909"}
{"id": "train_038082", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "3ded3d6757f871d8be45a900bea9d9f63658fac4"}
{"id": "train_038083", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "4702d42ad55190a02a872f2c36571d009af5a4bb"}
{"id": "train_038084", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "3fbbdac39644e5186ef1131ef21cad08332d196e"}
{"id": "train_038085", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "780227552b2d2f617100f1c5f7ec2115a80328b0"}
{"id": "train_038086", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "38c547184f396e1219360e5e04e16122a1578f54"}
{"id": "train_038087", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b35046027cb2275c4d681641020b85b1c7216140"}
{"id": "train_038088", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "8be4de57c505b41e69e1e7e06bf7c51725d1668d"}
{"id": "train_038089", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "4476809e6bbf860c673e245366d4f455967a9875"}
{"id": "train_038090", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "87785d212ff2d46fee7308e60f09ff2e2802a25f"}
{"id": "train_038091", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f269ad566856cf4447a2ffadfdc258677d70170d"}
{"id": "train_038092", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a833ef6c370ba2868014b667f04d7bda55acb56c"}
{"id": "train_038093", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e3798ffff1addacbf4d969261f8d1393b3a73cb0"}
{"id": "train_038094", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b1dff6f8c3e35b0f1b56c547bfc643e92e2aec7c"}
{"id": "train_038095", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "6da902051c46228746d1a42abe973e19c05a8e25"}
{"id": "train_038096", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "ac59ff758f563480600e2d00879c89367ed5e5d6"}
{"id": "train_038097", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "af9bbb0f5db31671a7b52f4f304822f528050826"}
{"id": "train_038098", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2bf84414f60b30dca744f770bfb5d92794761dd2"}
{"id": "train_038099", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "ad808461a87edd0a486e915749a623b67cc149e8"}
{"id": "train_038100", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "2e8af457e21795388f333636864ad2c08b095ea7"}
{"id": "train_038101", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a6e6bb42214aba0e44edfd532c420b686f2ba908"}
{"id": "train_038102", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "fcd2aa462d81e2ede7bb8d2d47159fa0c0b5758d"}
{"id": "train_038103", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "317e55377b0007d1ba1ed859923d798ea9303e66"}
{"id": "train_038104", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "79e9dbc8b60db978210af421f6f7dce5d0384dd9"}
{"id": "train_038105", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "06ec99375af12dca663a74182b2dfa731d7cbb4a"}
{"id": "train_038106", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "812865355dd7f880079da1d8a1d012d6ccfd4ee1"}
{"id": "train_038107", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "38a1d0503aded645e94c222b848f42fc11d88efb"}
{"id": "train_038108", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "1a3ed2b9f8c71abefea20eecbec7c1e8dbfe85db"}
{"id": "train_038109", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d4cc4fd29f8ab7dba7b501d978529e0551836759"}
{"id": "train_038110", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "7c3e651eb87161235296d7289d69defab8894e2e"}
{"id": "train_038111", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "47d3da19ce702c5c99552914e2ee51ca620f77a9"}
{"id": "train_038112", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "fe34d0657cb51016bee9e6e843d958cd3e313cfd"}
{"id": "train_038113", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "926ba2b157656f31a1e3ab79df42631a280d21e7"}
{"id": "train_038114", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "313a79c02ab96e81b338bd132d22ad4e3818dc7d"}
{"id": "train_038115", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7ec3e767fa720c68408ad54c8cfbb6b6604d6008"}
{"id": "train_038116", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "fb3c7310365cc4c9f25e00642bc7e4cefcdc89ed"}
{"id": "train_038117", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "cec73c47ec5b464774003f4b875a7471946e19dd"}
{"id": "train_038118", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "73bee0fb529af472335814a6d7ce09df713a5386"}
{"id": "train_038119", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "00660d5b4761c09a2f8c2f0ff898a580a2ed1181"}
{"id": "train_038120", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "1a9fc1ef956a9d0a14aa2f979cac50ed9697ce40"}
{"id": "train_038121", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2791f665838fae460710e97cc0fd6afb2974360f"}
{"id": "train_038122", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "1d9c14923cd46c12a1dfeb73ffb26facc3122e60"}
{"id": "train_038123", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c7fe978ab0487b355ba94ee092d071f58ad5462c"}
{"id": "train_038124", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4c75e54212bf9ff7ed670864925f54e2a272bd6e"}
{"id": "train_038125", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d105a868ff54134e4dd44c37cf96525918373541"}
{"id": "train_038126", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "720a711e8b2bd378e0a205dfe7977c061eb4ac38"}
{"id": "train_038127", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "eb109b535e5356db691f54d5564f7ebe5bf37d58"}
{"id": "train_038128", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "cbc6ab9b8ce23f36d0823365b2f902237dc3ff7a"}
{"id": "train_038129", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "a8aa03c92fdbc3bd07db1f702be3dc761a82b8ba"}
{"id": "train_038130", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "62adf74024e5ac2e0c5471fe6c2226a64185d93a"}
{"id": "train_038131", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e52929bcebaf5dc7be1ba0ec2e1e4c16cb127eca"}
{"id": "train_038132", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "37ac7aa69436224aa6729de955f6599781f38b59"}
{"id": "train_038133", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "410ddfaedf7a9b2448631791f089b54f456681b4"}
{"id": "train_038134", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a2f69b71141da58b0744f85f0873bf419c38adfe"}
{"id": "train_038135", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8b77fdb65dacef6a1a93219a045cc12ae6c9a009"}
{"id": "train_038136", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a51ca02b7fb10efaed85c40c77a35932282d14c4"}
{"id": "train_038137", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "06538d880d55a63d451fbcb96fb16b755f9d7161"}
{"id": "train_038138", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c6bb49038272974d4bfe7b12c7ccec6bf804c6dd"}
{"id": "train_038139", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a803b3cc280d03bbfbc7b0dc26756a52c577f40c"}
{"id": "train_038140", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b0af7c43b056d3ab8aa36d82a8602a3811095376"}
{"id": "train_038141", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "58787be06985ee26ed65f13e1f9303e03f5a5c34"}
{"id": "train_038142", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a8c786dd3779db0c838fd6bd94e5d9955f99fb6d"}
{"id": "train_038143", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "7adfcd1355ab7e6abb8e27f6e2da2e7356b94d65"}
{"id": "train_038144", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "b71a7ac755cd80dda3ae414a4162916043002960"}
{"id": "train_038145", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "ca003ebce55b8150c68d2cac1ece8e4831dd16cc"}
{"id": "train_038146", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7f0a8ea07740a6f5496f42d253e37231f5e1e9aa"}
{"id": "train_038147", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "4c8cda7d22abce6c24f28a478738bb9f786fd046"}
{"id": "train_038148", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "3ed4669dc7fd1afe50e06a441e9b390a86ad32f6"}
{"id": "train_038149", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "5c1f61cf0e63510899984de101720f2d47f69a4c"}
{"id": "train_038150", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "1430fb6b7ce7e2cb070f66ba05329cd6febdca7d"}
{"id": "train_038151", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "05497bb13a112241e43219156605a4296a93585d"}
{"id": "train_038152", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ecf68be4913f3de8b49f6add502a2e3dc4802d64"}
{"id": "train_038153", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8e0273ec161d7277b15db018b88aefa333297a20"}
{"id": "train_038154", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["governance", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "8165940eebc3729b93f5affc2a88393e91641712"}
{"id": "train_038155", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4e1cc371cbe60b740f8867fe32cf7f85caa303b0"}
{"id": "train_038156", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "9129a57d4685522d829bde37b2ad02667827d2c4"}
{"id": "train_038157", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "089b185bb3517c86de512b9609a336b34ced7eb5"}
{"id": "train_038158", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "2ac64879e7be3bc380b6f2dad84b63146267aad7"}
{"id": "train_038159", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c4f122a87d45e4fdace104961e940b067f896ab0"}
{"id": "train_038160", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e355ff449fe017b3a0e28112d13bc0f7e46432c2"}
{"id": "train_038161", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "02253f77f6c5abbb1fc7166ca98d7b0473ff9e9e"}
{"id": "train_038162", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "9a4f2cc7ec1795fa7a502bb94c88bde20d864d92"}
{"id": "train_038163", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d3b17a19b54270126d35310868ece6e5e5c64f4c"}
{"id": "train_038164", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "46b14ee149b7cf3d1a52846b9dfc4986884cc00a"}
{"id": "train_038165", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "b84f2083f0ba775eb50b4f0f42c633a030810ea0"}
{"id": "train_038166", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "392686273cd4400e7f504bbe3ec48063615c23ef"}
{"id": "train_038167", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "76731075d4eb4e5b69d996eb6f85f8787280b96e"}
{"id": "train_038168", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "b93812e2c50be8c3605cb16542358acbef6e7394"}
{"id": "train_038169", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "424565d7e8db39562040afb125af8f2731eff37e"}
{"id": "train_038170", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c5adb2b00c30fc675ed93882504f13a9bf83ecc9"}
{"id": "train_038171", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "5a960a4b58e1ea0513c560cea0f653c929f26c84"}
{"id": "train_038172", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "87760c18f3ee2c6e67167cdb2ed8853eb927c4a0"}
{"id": "train_038173", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "7ee534ca9c35ff8e731a4f1276e68d728d8d82bf"}
{"id": "train_038174", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "90919dd52e99a2a2b91703ffe382fdc5985fce21"}
{"id": "train_038175", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "890e6583cb1e9274e166e8a469a582ffc86cb56f"}
{"id": "train_038176", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "b3dd0ad993c51363b9578a808e1f6e037e0e50f4"}
{"id": "train_038177", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "04534c90b92cb9bb1d0b68fe48f33781a238d625"}
{"id": "train_038178", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "0dbf4ff665608cf8d54f10f7b6bfce308e7983a0"}
{"id": "train_038179", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "344563f5f359fdcae729d2bf32c9b6365a59b8db"}
{"id": "train_038180", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ee7d10fbb60764854446db025361f9a115f220db"}
{"id": "train_038181", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "24c2ccd5cfe97460623e91e63339f34e1575ecfd"}
{"id": "train_038182", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "bb3b166e234a26592d2800407e84e2fd4675d3ac"}
{"id": "train_038183", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "43703d8ad330110881c3a69e4d9cd1ee9a0faae0"}
{"id": "train_038184", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "1ffd95f8b88fac51c3adc92f6e64b3b4043cd84e"}
{"id": "train_038185", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b1bfe8e92afd10322f8cfbfb87ecc8cfa8d6cb40"}
{"id": "train_038186", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "44807b8d43f441a84590a3056a6b465426d93913"}
{"id": "train_038187", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a89c191e365f074e90645257bb1ebce4c8f0f8d1"}
{"id": "train_038188", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "9516a6c014d8ce61b308abc65416c4c26f07b732"}
{"id": "train_038189", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "bc319dcc138471c8dba646c25d7fce7c1f01218c"}
{"id": "train_038190", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "bc7d7a9e4d8bdb2ee98f76fc0e391097c6db42ed"}
{"id": "train_038191", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "cd644527648db31004ce1cba5067dc2e36b6ff8e"}
{"id": "train_038192", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "25e17c2d65f9697bc2f135265df9c19614fdcab5"}
{"id": "train_038193", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "cbc3236cc05247b12e20d155083dc3a23c28fe71"}
{"id": "train_038194", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "725825cf1f9fd5a6d3e40b72b911189ab78c58a9"}
{"id": "train_038195", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b18fe094bef8eb9ff67d0be286f123613c0a3041"}
{"id": "train_038196", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "87b6049928f09fa66c82b8b5f484756a3ffad1fd"}
{"id": "train_038197", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "9d543fdc94add9b153240a7d87e34e59247dd992"}
{"id": "train_038198", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "3c5d63ce00da209a4457229aef831fa492bd73e4"}
{"id": "train_038199", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "0795eedce0ca24105dc28bb0a5659ae949aecf45"}
{"id": "train_038200", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "097a054c3e66ba3b74230a252f06549c6df9eb8f"}
{"id": "train_038201", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ddd00753b278e5fa5125f5065a4a4d78da0da3bb"}
{"id": "train_038202", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "4982a5c546747167171e6cac20a0485a6acfad00"}
{"id": "train_038203", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "975fe26be668ed6d224566dd9057e4d812467e23"}
{"id": "train_038204", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "2db933f4e1d3ad3e1dd0a11363daeef27ba9bd2c"}
{"id": "train_038205", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["agent", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "bdea4e5385fc214fbdfa24a582b207377eab9bfc"}
{"id": "train_038206", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "626fdab58f2e27525f312243ff24f91c2eebf3bc"}
{"id": "train_038207", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b0dc8b40a07986ba9146abe3fd771d212b8a7424"}
{"id": "train_038208", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "3394719c4a32fec74503c8a4b2b41e6bf02743ac"}
{"id": "train_038209", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "1147772c6f8a2b7dda32cb5057f2ec84a1aac007"}
{"id": "train_038210", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "198908d6a4f73e8fefeefd0d6da46c20d2824c98"}
{"id": "train_038211", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "f4e290e1cdce44c4bff8efcf81b0851fd2055a80"}
{"id": "train_038212", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "742352027cf265e7d42b451570f28d3ddf655b11"}
{"id": "train_038213", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "70e9da4dc29ab8c04ace2c904db4c32a918759ed"}
{"id": "train_038214", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c8f5f75f43c48f85f3620c3b344f0236b1a3b5b1"}
{"id": "train_038215", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "7f0a8ea07740a6f5496f42d253e37231f5e1e9aa"}
{"id": "train_038216", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "5e93fbba6e575fc5894214c03785f59d424227a8"}
{"id": "train_038217", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "063c1f4cf1017426b55d8daba5520848e58a1220"}
{"id": "train_038218", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c970982f979f3086122a23d78cd6703a7cdd71d4"}
{"id": "train_038219", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "2e6737f76bbd8dac7ffb58f9604721f6333856e4"}
{"id": "train_038220", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d661555dfb969aefbf9674f386755a3cde19d49e"}
{"id": "train_038221", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "be9f130cb86a4a3c304a86d474be39e62de70b8b"}
{"id": "train_038222", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "b5d43be0ba36e39c0073517a7c67a738cebb2d27"}
{"id": "train_038223", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "2e932b57a5369934d2c93f012fae15fc5dfcb4a0"}
{"id": "train_038224", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "235a796ed628f47db4684d117e3f0115e916cbd3"}
{"id": "train_038225", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "eae8875621837df62953057dfddf9dfeaa05216a"}
{"id": "train_038226", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "cf02f5cb4cc6a04ff5b7c34cb4c08aeab7174a55"}
{"id": "train_038227", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "689fb0b45e2dd13167b639de9967cdd5d543d77d"}
{"id": "train_038228", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "321072ec108a0475034a422862d90f9afd4d5559"}
{"id": "train_038229", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "62caa9e90e28e776a3ec84f065aa7d54999bd916"}
{"id": "train_038230", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "b24a6cd1c9da599fa37c4f965aebab56e074298e"}
{"id": "train_038231", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "621faea85ff1691009f2b655a82371939da630cb"}
{"id": "train_038232", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ff749d7e95c9b6589821512888315f179cec4f01"}
{"id": "train_038233", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "783b3cb29716f883bb0282502c80cd8246dccc9a"}
{"id": "train_038234", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "b1654f03f14dac7a86613e0f6704515a254993c9"}
{"id": "train_038235", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e98fc399dc21d97403a11692f57bd07df775ff43"}
{"id": "train_038236", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "3124eb984ba349ad49719d6beea7f4874befe4d1"}
{"id": "train_038237", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "bfda0c735509329ee19f2c10f0a3dd317aa6fe1c"}
{"id": "train_038238", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "8b8b0f9d3f10eda50e9255863779eb215e05983a"}
{"id": "train_038239", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "15cf041d06c660448bb3fc6bfd090ba0be1e9a9a"}
{"id": "train_038240", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "3a14f4bf37a10ef2988d485364192a13b47e93a8"}
{"id": "train_038241", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "918f0f86373f55a1ea3773f5867b398bc4b97c5b"}
{"id": "train_038242", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "aaec72989924dd3cee917f263d021909273cce20"}
{"id": "train_038243", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "27af865359880c5cca6022bdfd42c346f6e25f6e"}
{"id": "train_038244", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "27598e1961e5b310d1525f5faf6a95de2b87670c"}
{"id": "train_038245", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "871ed461cc539f29d9484b78ab8b96dacc10b45a"}
{"id": "train_038246", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "4286e72b6e7cb1a67ca110a84890480c05f1ceeb"}
{"id": "train_038247", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "891d31097b4671506713ae352e3f98b65f4f238d"}
{"id": "train_038248", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "9ee08b3c4c64ecef985eaf02947457cf9823e6ee"}
{"id": "train_038249", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "bd832b72be7a1f956430959e12430a818425e003"}
{"id": "train_038250", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0c54fb358d2d838953e8243ab38acac79442e5eb"}
{"id": "train_038251", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e55896dc9aa4e86d51b0a092cc1df5ca0bf1af55"}
{"id": "train_038252", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "39dec78c398880d764f30d3f0cae74d684682f30"}
{"id": "train_038253", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "672779d1c363f5a05e68211522d4143faeedb3ba"}
{"id": "train_038254", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "ec9a59af0596bd9fb562240c269c1b60b1e02561"}
{"id": "train_038255", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f10cfbcb5cd918cb5129f164043b23383e967927"}
{"id": "train_038256", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6d83d3cd80c3cd3b14a0571d5404f1148b34ffc7"}
{"id": "train_038257", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "e7e0ebd7b0abb32d8bc969d9d1a3ca4ba206e3ec"}
{"id": "train_038258", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["governance", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f230463b196dfc7a29234a13cf4d0c2fd742b75a"}
{"id": "train_038259", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "b189f6c91bdf64fe9eb706782871b54758275e6f"}
{"id": "train_038260", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "1887bc1dd5a3fc81df895c580d78e43842132201"}
{"id": "train_038261", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "424565d7e8db39562040afb125af8f2731eff37e"}
{"id": "train_038262", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "a40733a05ba08c8c0208a36c94b59b65a0a0fb1f"}
{"id": "train_038263", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "249cfb91c5ec2716db868ee6c2aff11da66dde6f"}
{"id": "train_038264", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "4a773d76b67466d487f0590c32212a2f834bde68"}
{"id": "train_038265", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ac076700102a0899ecd83185aae709428e53a82d"}
{"id": "train_038266", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "14fbaecd2d21565b5551871163a646ea53ab2686"}
{"id": "train_038267", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "415c45580a50729e670118925be44d1cb7336c5d"}
{"id": "train_038268", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "0f28acc762ef89a1951cf2454b54324cd8cc583c"}
{"id": "train_038269", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "075c2ad7c4254d5d109e95b43eda07d9a9c59f66"}
{"id": "train_038270", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d04b0cb2ef777ce2a2c44ba80c4b725827ed6254"}
{"id": "train_038271", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "3a040a15f2d024244458de24a7ae966a0eb451d9"}
{"id": "train_038272", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8ad98081bf157e34a7696cc2234332ca4d67073b"}
{"id": "train_038273", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "016ea8bbeb9d340981e41cc3c7721edc699b14b4"}
{"id": "train_038274", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "06fabb38ef8a49bd6e82253cc35381622a596262"}
{"id": "train_038275", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4259b3fa7b3e8413324fa7eec547b6eaea9a0d8e"}
{"id": "train_038276", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "0d6dd70c455a8576312efe75aae4c323b5d90f0c"}
{"id": "train_038277", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b69b70f91ca86eb798ab13774dde60b835043c32"}
{"id": "train_038278", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "0dc0f561adc6e393859e162cb6c8991c7368027f"}
{"id": "train_038279", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0660f3976174cf5359d0b8fb26a02290243ef8aa"}
{"id": "train_038280", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d0d328e5071eb73e44d7573a542ab5180d003337"}
{"id": "train_038281", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "02488ee3bec61f25f8982a6a77b26c1235bc6db9"}
{"id": "train_038282", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "727d7e678b70de6fd498de2c824902b472a04325"}
{"id": "train_038283", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "39c757fb9e79aed888e50311497400b05d3145ac"}
{"id": "train_038284", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6e9d2d15859d8e3fb08895e6827842c3f60dabb4"}
{"id": "train_038285", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "cce81ba46fe0ffd603ec76385f4d07a244b219b2"}
{"id": "train_038286", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "61d4c6951d2a7eae5552a6f697ca729a115b7d7e"}
{"id": "train_038287", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "daae77428df578ad5cfc88980c6fb37917e652dc"}
{"id": "train_038288", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "8f90e38d9ecd9a619457cedcb216406f7d39d63b"}
{"id": "train_038289", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7915cc497abff575246d1e3256dd1fbe93df8b8b"}
{"id": "train_038290", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "c4e706d786de232a553a360278941762e04e1404"}
{"id": "train_038291", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "f29c2d364cbf84a2b9b96d87f340c6b5f2d9c3ac"}
{"id": "train_038292", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "8b7f6f40cf6dc1855ffb97d291996d48dfb06e71"}
{"id": "train_038293", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b20b0c2f2285741aa1532713764ffc5cab1a9a28"}
{"id": "train_038294", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "da6371b5c5440d96a5a2c88d454f406685902d94"}
{"id": "train_038295", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "d8d55ef052b899db454afc4c8cb937222729ec1b"}
{"id": "train_038296", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f112bc20c5d9a8eb3bc6549d0a6eaf265dff6608"}
{"id": "train_038297", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2ca6310a8e70e1f721611a5354362e9add1e1748"}
{"id": "train_038298", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "39c2c5d3595e05e8c1387e70791ee796c1258df0"}
{"id": "train_038299", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "f0341b21544a8975e8b6f62bc199310fb2a2a7d3"}
{"id": "train_038300", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f42f0c350677cbedba664d6d8cb31d65b43f4eb9"}
{"id": "train_038301", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "11930b9fd87838c39fe15c14cdcf4f6a46f214fd"}
{"id": "train_038302", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "54264bf7a35c2dc79269601a5f3677fb8bad6250"}
{"id": "train_038303", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "865563eff409c09392fda94e5cbe62c31334a310"}
{"id": "train_038304", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "9e05c87417f5b55872e9f423b1c8b6fa76401a0e"}
{"id": "train_038305", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "86cb68e54940678c9072aabcbb125a69ab60b822"}
{"id": "train_038306", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "d279e8aa68ea7b585b9b7839fac97eab059cb9f2"}
{"id": "train_038307", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d62430c0494ea0a9dd7c28b389314610a32bb7ea"}
{"id": "train_038308", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "b91c60a25f58cd216d4b11b4bb6f83a026b47ef8"}
{"id": "train_038309", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "928fb72d3e8cb6b1683e16cab258df608e8d27a1"}
{"id": "train_038310", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "847559d499bfc14eb3998a167875dbe6175e6fca"}
{"id": "train_038311", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "734a96bfcbd42b24f6f27ae8971f20ea67f39cd9"}
{"id": "train_038312", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c9c3777b41400cf16c416f952b36f1cfbcf37302"}
{"id": "train_038313", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f4ad8e7b329e44edf63be41f19fcc96c99355e70"}
{"id": "train_038314", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "436eab3c3a47bbee290147f1a78fb40c01373868"}
{"id": "train_038315", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7bde3290ab656dd7dbace389ca1bc167699060c7"}
{"id": "train_038316", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "a5f05ab1f421cba1d5fa60b07bd2901180763af7"}
{"id": "train_038317", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "f1af1cff96f9bef9c278709214ddbde817b4bb1d"}
{"id": "train_038318", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "12e78e160631b3ea2eb78a4ede0fa09d28d50d02"}
{"id": "train_038319", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b84f2083f0ba775eb50b4f0f42c633a030810ea0"}
{"id": "train_038320", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "6c134b4f3f5d52ca6165193846cf179107216af4"}
{"id": "train_038321", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "10d1f54406c8953d4bce83567625321b63bbc46d"}
{"id": "train_038322", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ccca1c2eba7ab99baa1b40a56d5fb51096769cb6"}
{"id": "train_038323", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "cc204a9de94a6cda6b31817472ad572941bff2f0"}
{"id": "train_038324", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "893b6e0cfcbe5ae5ceb27a2eef92f84352a89732"}
{"id": "train_038325", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "693dcacd15eea421e81010241a32d4f0c86c14a4"}
{"id": "train_038326", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "32fe57a52898420c93fbc913862d25db87cddff1"}
{"id": "train_038327", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "03cfabc56e81faa5c08e76901c58f7312dc2d5d4"}
{"id": "train_038328", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0804e2b92e03bdf5f6e6d3101c94e0e448f7c073"}
{"id": "train_038329", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "a1e03d118111cc2a72b39e718a39fa4e1b8c5819"}
{"id": "train_038330", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "144fce9834737c201451823caa0aa28cccfdc818"}
{"id": "train_038331", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "9c713a249236eb9a02443e9c1e800b2d6634acf2"}
{"id": "train_038332", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "a10f823d023cb39fd5d48cd770d54d72ebaf8309"}
{"id": "train_038333", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "8c3ac9cc2f58e0275f5c52dee2a566cdc7b148ce"}
{"id": "train_038334", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "3ee04c0a914b962ef66485db9f1de5f7178ad63b"}
{"id": "train_038335", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "3f0e32b08b7ebbd1f8389c00f88da9a793934cb2"}
{"id": "train_038336", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "7820aca5fa808950243a32a510b4c543c60eb201"}
{"id": "train_038337", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "e07d0764a6e359e1049497d39e1dbdf33f011d32"}
{"id": "train_038338", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "0891825b5ce31a55e807ae9d5c4e2332bb25dd7a"}
{"id": "train_038339", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "15886ea4b8fbce078762bde40bd2145ef1100a53"}
{"id": "train_038340", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "39fd641b58200730740c649fd6e617f9b3272315"}
{"id": "train_038341", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "dadac65941a0a6261592fe38c25ac631644bcc20"}
{"id": "train_038342", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a45de7f9867b5afd1482c8dde49d874fd01e73bd"}
{"id": "train_038343", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e9bcc0a33b1d957528bfd6c5fbc5a4fc536d77a2"}
{"id": "train_038344", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "cb05d564f8b2a58cffb61f1c122c0776c2720946"}
{"id": "train_038345", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "6f6b84b5285df474e47d01a1b66b623ebf43ec05"}
{"id": "train_038346", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "29bb2c67b00af5482a469be40ddf8b3944975a25"}
{"id": "train_038347", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "985aa077c72f30f636fbc021536407463f1eea35"}
{"id": "train_038348", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "2eaaaae44c102ea9634ff54be224c696ec1348d0"}
{"id": "train_038349", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e619f1a65303b4900e364b5f09182dc2c1eb9ee7"}
{"id": "train_038350", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "08f3a4e9318521860f581b2ce73a56f26c128812"}
{"id": "train_038351", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c88967ea78642e624d04dda07ab246c77c884338"}
{"id": "train_038352", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "99f3b7ddecd75016c8bbe52c59e031b183864b25"}
{"id": "train_038353", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f1b160bd15f08ffa3328092e5776f31afba73d12"}
{"id": "train_038354", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d6951fd8cab92fd01197bb0d98b9181642e067e9"}
{"id": "train_038355", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "90d475b4c1bda25fb8302e524f7db9a50dd7fe32"}
{"id": "train_038356", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "4e9ae0bfacdd5c3dae82ac684b46a21f70bdde46"}
{"id": "train_038357", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "a979930e9dcc827e30f52503ab51432c63f20a97"}
{"id": "train_038358", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "bf4117d0f42a525689c1ab5208720ce255d6a4ee"}
{"id": "train_038359", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "9faaea85eb91c55187d66811676ce4e9e391ea5f"}
{"id": "train_038360", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "2a7eb3f67411dbd00006f8cccee4bd7a9bfb94a8"}
{"id": "train_038361", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "0b850ceb11934bb0dba96678a6cb17947a547a8b"}
{"id": "train_038362", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e0980b656848d6d02c2610f1c0d214d5cef7e625"}
{"id": "train_038363", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ccbb5c2949b17839947b7b48cf66bbb2f527f951"}
{"id": "train_038364", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f3a5e966661b18bce6b8d0b0b36b1de2b2113904"}
{"id": "train_038365", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3e8f1e714f5275b8fdaab2892ecd303bc41e1944"}
{"id": "train_038366", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "00ce63bb9aa0d9458ae0313ea987a06c924dbd99"}
{"id": "train_038367", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "5d5595d0e6b1edf0b6cacab9d534d876a72b8513"}
{"id": "train_038368", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "ddb5a4aba9740c8831b435d7313cf7904ee6ac9d"}
{"id": "train_038369", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "63892054bb80f18979ff1db2f0391538fcd8fe79"}
{"id": "train_038370", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "65ea11a2581fb250fd911713429fdc924bc68ebe"}
{"id": "train_038371", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4dc9878093daf2956d3145ac4cd671809731651b"}
{"id": "train_038372", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "4d2776efd72e12a145c5a382e9eaee610cbbc2a8"}
{"id": "train_038373", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "1abe2b2c056152d8d4ba6a23c651e76b383593b9"}
{"id": "train_038374", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2c59bdf91f4852a1980c59de0976722821c1e1fa"}
{"id": "train_038375", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "ec8ae4355e38d39b82e6ca03358199771b84d488"}
{"id": "train_038376", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "02172eb5e6ee2cc5ae142eff462683a5b4767eb7"}
{"id": "train_038377", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "2d4770679b800bc593313b5ed7f28d20aa492228"}
{"id": "train_038378", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "1b1ab963564cfed760588713325bcb1fbb79d87a"}
{"id": "train_038379", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "7c13398cec1d09748bcd17faada4299581fcb87b"}
{"id": "train_038380", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "3dbaf52f082f312be713e01c7ffa637b99bf9802"}
{"id": "train_038381", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "2b6eae4773e1df84b2c5f554811212ddb5229123"}
{"id": "train_038382", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "26af262c46fe82c80a7e85a125d68726d3cdd427"}
{"id": "train_038383", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "79c707a35c6945cd43057ea5f61549f593080a58"}
{"id": "train_038384", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6a73fab2b19d973f34fa6cdd6e00e86c64851dc3"}
{"id": "train_038385", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "8d6661bff4a8bf46ba54e6ec976bed4eb7e908fc"}
{"id": "train_038386", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "cd8210f0eadcf3f1c24e47811b60940aa16c2138"}
{"id": "train_038387", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "6c376405d100321d3904141d87d0b821e617ffad"}
{"id": "train_038388", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "5006fe2d4255162b7db3fa7befe3f3f4732a21b4"}
{"id": "train_038389", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2b370583451cf04a53bb8ac3305b8af80d1c76f0"}
{"id": "train_038390", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5e0b984f8ccc886feb82db8c1137057137e27074"}
{"id": "train_038391", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6b1dd74db0d54dfed2fe038ce29fa534fc501ca0"}
{"id": "train_038392", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "82660937cd7b51c93f56e75570be2f531535d164"}
{"id": "train_038393", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "805f3629596a6871dd5685eed7c40408e3688428"}
{"id": "train_038394", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c8dd13cc3468b27b737a9f38c07348cf1abc995c"}
{"id": "train_038395", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "2b63030313c76975c57428b71ea235c9d5910da2"}
{"id": "train_038396", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f4a62b5f6102b83ac9597179fe88ce987a87e0a4"}
{"id": "train_038397", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b325d27cc9ce8d8c718e8d692bdbc08491275fdb"}
{"id": "train_038398", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "2ec29f8a21e1edbdf549cab42e1b87f033f4345c"}
{"id": "train_038399", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "32c63f69bf9a5dee88199dc4889c6b73f8733786"}
{"id": "train_038400", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "db609ff6b691197cca6c7eab90113f3885a08bd2"}
{"id": "train_038401", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "5ede0e88a6da11dd0bf618a6accf61c1f3422c4c"}
{"id": "train_038402", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "0d2801cff61227ce6f96c63e1f9f9ea8ace971ab"}
{"id": "train_038403", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "559a01edd34635576b5305368d0d64beba01ee62"}
{"id": "train_038404", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5f13dfd7f541fb2617f765eb0d8aad66fc091e01"}
{"id": "train_038405", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b3decc8c291975e1c84df6d12270e6db18300a47"}
{"id": "train_038406", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f658721074960b475f1542647e79fdc03d20b127"}
{"id": "train_038407", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "2b78ae7e41d080b2f0357f6da73b1dc9f80fc7dc"}
{"id": "train_038408", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "6c4e353e6f8e469c552dd2190d9d710b44a3ab9c"}
{"id": "train_038409", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "825b735ffe52fb51700b34aea03a2486e314132b"}
{"id": "train_038410", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b66d9c01ea694b985d632ca5ca7609231e09f46c"}
{"id": "train_038411", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "965ee575acd2c8994b7cbdec9387b0eedef62285"}
{"id": "train_038412", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e8f72d24199833e951359b8ee5a65bddaca79b1c"}
{"id": "train_038413", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "71a8faa1781fc6fe12ccba584169dca75d5eb286"}
{"id": "train_038414", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "27e32673cf36ccc5b627afd77d5de16939095ed6"}
{"id": "train_038415", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ffc57ae6121b98e9597999367d472a2be1da93ae"}
{"id": "train_038416", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "cf3c877eba8979262da591a7e4bae64bf7a74d9c"}
{"id": "train_038417", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "b3c7564f745b9cd69cbb72c290aee29c8675f996"}
{"id": "train_038418", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "ba47cf720cb790907620a804e2edde7c9d41b66c"}
{"id": "train_038419", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "32dfac31fed7b1d63a1c0a52ba6df465e10d74db"}
{"id": "train_038420", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "17dd8ebf2cbc82547759a8fd200f5c0eda5a2368"}
{"id": "train_038421", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f6d44dd8131f1f852e57c6220fdb5bedaf2f8fc1"}
{"id": "train_038422", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "ab945bb8735aceaf1f4955391676482367cef1ec"}
{"id": "train_038423", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "e1d192fe7651d01102e15877949fe1a2cc0de101"}
{"id": "train_038424", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "3e870cfd2c5b6cf5ffc78da73aa34b14b79afbda"}
{"id": "train_038425", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "bdbf1a03465b101f5ef2eae20e33a3e0906ae4de"}
{"id": "train_038426", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "285cd96c32351ec304fe34fe29d61752d418699d"}
{"id": "train_038427", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "dfab1422c1e0dd36b75c344dcd4190685fc8427e"}
{"id": "train_038428", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "cb7628838ad56508eabb38bb56cf23cbac19d9ce"}
{"id": "train_038429", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "1f78697c18566e12753469554cb542e747a7e674"}
{"id": "train_038430", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4daec53665aada658780393e63c1f85e1a0f82e9"}
{"id": "train_038431", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "06541108438ea7a0b388c06ab61d8cb9ccd8faa5"}
{"id": "train_038432", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "b556cadf8bd1454d7f9e59ecfd3ed809da07badb"}
{"id": "train_038433", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5708631620e2e0c4f7b72d04528de7f725aeb545"}
{"id": "train_038434", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "a75f4b4df1ff65f5b16ef40dd5b1a7857c9ea779"}
{"id": "train_038435", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "9721513e07022f47ae2a28d968a2b56d8f8426dd"}
{"id": "train_038436", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "3ade521cd812729c68cdbb884d997d533d5dc91c"}
{"id": "train_038437", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a47673ca5186e1067c999b7d49df71fc2dda1dc8"}
{"id": "train_038438", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "272965b56b30d1605b11013ebc2aa77df678b2a4"}
{"id": "train_038439", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "b1e3c222dcebc4aeab23adba66a756bac2c5fec9"}
{"id": "train_038440", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "5fe97ea47e593d468bba4291d991876bce87a73b"}
{"id": "train_038441", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d395a6dc6b5e0dcf2436776bacc2d367812ca24a"}
{"id": "train_038442", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c20dea5f13661d0e8690f5a668867b9a09813ac8"}
{"id": "train_038443", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "cc8b5557247833cf51ca4cd23db688ef32d66370"}
{"id": "train_038444", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "5cc20219451136e25e60d83a7132284643faf250"}
{"id": "train_038445", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ed25b1e696fe2ac63acefe4110849444703b5541"}
{"id": "train_038446", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "da6afa1f3e7cd8bf518e4a0374d69756189ae8e0"}
{"id": "train_038447", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "47c44788dfe583927f292581df4759f329729a89"}
{"id": "train_038448", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "02985baab6d3c15f6e9a71e91c28da93f118d4a6"}
{"id": "train_038449", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "25a030436759f9749221c843665e95566b4dd6a1"}
{"id": "train_038450", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "7c8f79d4ebb6a3a1c3fe06577ae25de3293172f1"}
{"id": "train_038451", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "941eec5767a3a9f2dea65594590424717ac3c276"}
{"id": "train_038452", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "80c0a9bb891f74dc67d6d4ffbed942280bcbd9c7"}
{"id": "train_038453", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "1d993443d75a272b4719925bba480c61a199a576"}
{"id": "train_038454", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "438dabaa2dde99ab9225ac826816d34955ea6099"}
{"id": "train_038455", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "5d87fb16f69bbbbe71fcce5fcd49f8b177253c74"}
{"id": "train_038456", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "0f7f3624deadc681c06547d970e4adb5e41fcb31"}
{"id": "train_038457", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "5259f27968afe07c1d95d99ebcc37a234134f9ef"}
{"id": "train_038458", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "7a56e9619dc2b5b0749b8627983e35f2c9b26d89"}
{"id": "train_038459", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "9788c26098279184d8e586dc28024de2d50ffe52"}
{"id": "train_038460", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "ff16ee4a422c26b930ca1947de3229e18be4e9f4"}
{"id": "train_038461", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6cdcefbe108298c3adf26fcab01d1bbaaac67493"}
{"id": "train_038462", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2ff5312f4c95332ccff4a35c01d793e45d4a9fc3"}
{"id": "train_038463", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f6807152e62e6e511a9152b928a1f2cbb3dec0f6"}
{"id": "train_038464", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "38afbf1f77e820baff489d13a1af4b59850a6e73"}
{"id": "train_038465", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c5c696b108f53ab792b44dde3d94aa0e77383609"}
{"id": "train_038466", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "5aa272da6c06ce0bf9562041bdcdae44661424fa"}
{"id": "train_038467", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "be68a1e3f4c7c35d319b0f7d96c8101b07d07d26"}
{"id": "train_038468", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7a44f05fb25f88b1cd133ba450726d79a2a7c80b"}
{"id": "train_038469", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "61bea16932c7c5b6cbcbe258971b0488909c6492"}
{"id": "train_038470", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "7d13043aa55abfc83f4ad14f2e70f3deab074321"}
{"id": "train_038471", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c8dd13cc3468b27b737a9f38c07348cf1abc995c"}
{"id": "train_038472", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "090a02735d49737467423ce1708c1664111adc9a"}
{"id": "train_038473", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6edc00799b28e0ca20c570020a4f93f855658bf4"}
{"id": "train_038474", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "4de9016f549e5e884a6496d2d83761f890a58efe"}
{"id": "train_038475", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "36ae91617ca72d828427514769cd69724fa881e9"}
{"id": "train_038476", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "7307b3ac3bd6ca3aae5f5647dcf361ae912eb19c"}
{"id": "train_038477", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "1db73beeefa7999e70b6c8fa6d36a5208ce7769f"}
{"id": "train_038478", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["evaluation", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "33826659dc6e0e5fafeac01234975aa0b8531d6f"}
{"id": "train_038479", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "f6d24548d6f1aa55c5165685cba728c3807e0fb3"}
{"id": "train_038480", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "640fde0d99d066e8a4a728e9f1745b922984ab8e"}
{"id": "train_038481", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "87790017ec8db8bd43a32a6c0cab80936a454bb4"}
{"id": "train_038482", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "079047e1ab7bc10ee7be630eb47cdfa4e7b59b94"}
{"id": "train_038483", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "772cba8a127e04fe8d2c670a9d6a716b70971ca0"}
{"id": "train_038484", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d1d13a07ec3cb1dccc57be1026ffd44a98fff5f1"}
{"id": "train_038485", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "5de52ddf3194776a11dc0c21238af88edda93100"}
{"id": "train_038486", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "1cc71091134498597c0104778ee74c5256f72e86"}
{"id": "train_038487", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "aaf551655bee423308396ad2869b6636bcdf9edd"}
{"id": "train_038488", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "019cfde6172121ecf696dbb86c555a4f191b6d2c"}
{"id": "train_038489", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "fa15d388d7e9447fd9f4806a1464ad0d2ab397a6"}
{"id": "train_038490", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "48015c442e2725da3fb93bd03954c1147470ba9f"}
{"id": "train_038491", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c1f78857508df826592158d0e9553f6ea66f9d67"}
{"id": "train_038492", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "28bd1371f049721c0872a9731989a725d094aa5f"}
{"id": "train_038493", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2304fa5507f5e8fc3754dfeb47a5829ebcd14bbe"}
{"id": "train_038494", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "9f1dd16e8a00bc77444d33ba406cf8d6fbfa61c0"}
{"id": "train_038495", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "6243243945e8b1e2692bfb6c268b36c2e72c79b8"}
{"id": "train_038496", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "287a3d1afa6c9d0d627cd0b6d3f346f84819ce7e"}
{"id": "train_038497", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "62559381f0710a8b84e291027d089144ee0b5880"}
{"id": "train_038498", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "cfdd9a8946f663679322e61737a271a27e91108e"}
{"id": "train_038499", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "f5ba595ba17a588d39017f76f7619e01ca760f75"}
{"id": "train_038500", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "fc354ebfe8f0d514f9a3eb5fca7b8105f2e13666"}
{"id": "train_038501", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "9825428d21278b1db8c21b46da390ffed8eeffdc"}
{"id": "train_038502", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "571cc2a01387255b92012fe0035f19da66780c31"}
{"id": "train_038503", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b1318264e23e1866ae3b22679cfad3cf656bd413"}
{"id": "train_038504", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["evaluation", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a42b138983a384ea4ef9327b4b687732eb2b9440"}
{"id": "train_038505", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "289dd6997f901a6985c2cb8686a54a76ca2e0f33"}
{"id": "train_038506", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "f84e00a4a7552e943c6cc94e1f5c3ab410709374"}
{"id": "train_038507", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "9f51547a54c300e78f3576364e88cd5f297b22e5"}
{"id": "train_038508", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "307954f4f83624ce595b70725bc332f5a40066b9"}
{"id": "train_038509", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b1a1463c0ef9ddfa0d264e72df910b5871434693"}
{"id": "train_038510", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "55182e13f4418d262e15f771ee2a6862ee75e54c"}
{"id": "train_038511", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "725825cf1f9fd5a6d3e40b72b911189ab78c58a9"}
{"id": "train_038512", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "94761ab214640dfac1099f1c6498edaaaaec7c08"}
{"id": "train_038513", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "8a22d1b96bbd5d8b9167dcc39f2dacfc50bf66a0"}
{"id": "train_038514", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7619e4b6623f2208e711130278a99b5e5f23a023"}
{"id": "train_038515", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "65f6427688a267757c107e354f776d4d37f8a56e"}
{"id": "train_038516", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "8231ed40c212f25b1f3079c057477b52576c1fd7"}
{"id": "train_038517", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "a34dc14587e9af5c85b454e0a769a87ce534c653"}
{"id": "train_038518", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e4bdf19152541a61346c1354bc6c157d8d98f579"}
{"id": "train_038519", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e03beb3bc92b725f79d9e579da291181409f292a"}
{"id": "train_038520", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "600b2bef1a0e91652c5ba359367c6699da42c1ec"}
{"id": "train_038521", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "7434cd007340c5c18fa6e18e63c42e26f7864bf6"}
{"id": "train_038522", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "12c8dd0fd58e8b3421f5cd4eb87bc4865f9273a9"}
{"id": "train_038523", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "76f9178df81388077cd1d5e759c4ba19d795c628"}
{"id": "train_038524", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "cad4ea5bfe863221fb55d2a13367f683b8c83eca"}
{"id": "train_038525", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "0428655c1464015e8185e714dbca4c033e59f43c"}
{"id": "train_038526", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "3ac4769efa8572b39a9470484ad407e44b4a4090"}
{"id": "train_038527", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "b1b58282f08dcb47fabf925b90ef628a2b31e3f4"}
{"id": "train_038528", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "171579cfce59abe5786fbd7e983226ddf647a9c6"}
{"id": "train_038529", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "02b2c79eeae9e0b3873619d02a7e0f76cb103819"}
{"id": "train_038530", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "9bd36220018cc4033caa6b6e367f486e2a328037"}
{"id": "train_038531", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "41f203c2e54f4ae75582ac280d79380e96134a31"}
{"id": "train_038532", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "11443e12a7726f5af8b9a8c0de563aa196cfe379"}
{"id": "train_038533", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "9507e255e737133cd6cbd5baba07ecf9f587d23e"}
{"id": "train_038534", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0edea10f4c8e3772490accd00d2fdc319454bfbe"}
{"id": "train_038535", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "bd25d1bb807db30e2fde4c8c2e33a98906310155"}
{"id": "train_038536", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "852c0c1adc0bda9811a56f045ed7d40dcc0171be"}
{"id": "train_038537", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "0d6f9765b53ab90cf281a231abc6122723585834"}
{"id": "train_038538", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "2482d5b598fdb08f840e982ea71c5937f1c483d7"}
{"id": "train_038539", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "9483bc75c1c0dfe588f9c51b83f0010bccdb8093"}
{"id": "train_038540", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c1c05ce98280cada21015e02c96f605100365c63"}
{"id": "train_038541", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "cbbc23f1d89d0f0387729a4f6ea3031714dc6f40"}
{"id": "train_038542", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "daafa227f93f8888c4ddcda4927706e42a56ab73"}
{"id": "train_038543", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "defde792b548187063df2c672ef4f4cb67d026b6"}
{"id": "train_038544", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ae1875a74e9c2cd933835e6a793801d3dc17412a"}
{"id": "train_038545", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "fefd1408d4ac60ce760d65e5decb4fd0faf714e2"}
{"id": "train_038546", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d23fc2284ecb6797037a89bea17bc4ce41f2135d"}
{"id": "train_038547", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0d844f389887bb25a1055edc10a0244b805695f0"}
{"id": "train_038548", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "fc47443872fb8d3aea95d545669ebbb674f2659f"}
{"id": "train_038549", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "1a39e70b7337dbfbec8032ad11f24fe98560d773"}
{"id": "train_038550", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "17ce86531359bcd339eb7d4ca7f5fc71a7a46417"}
{"id": "train_038551", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2531a0dcba53d875d010ea5a11831bbf41915648"}
{"id": "train_038552", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "5e4c413e810760aa3136875e92f1442087f8e103"}
{"id": "train_038553", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "93ceea3a0493136fb4060186775df03f31231d1d"}
{"id": "train_038554", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "8dffbdbf7dc8a6e8c5becd4ac412f0bb595c436e"}
{"id": "train_038555", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "06941dc6f66334fd97678ce761009afd852ab9d3"}
{"id": "train_038556", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "1887bc1dd5a3fc81df895c580d78e43842132201"}
{"id": "train_038557", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "52ee8812434424a43de894bbb892aadf5fea7e02"}
{"id": "train_038558", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "b7c5cdd98fdcf34f0719ce5852f09ad42e72bc09"}
{"id": "train_038559", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "711a5a0e283622bdbb5fd76ad6246c1c2d681144"}
{"id": "train_038560", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "987e2e0a1caf514ad0972523a60dbace71a3aa2e"}
{"id": "train_038561", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "3089ba7cf5ad8ef38245f2f62f8463c928ee699f"}
{"id": "train_038562", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "df97888e042c97f257bdf3a4c935d0d5cb326036"}
{"id": "train_038563", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "443b5b1f9abdae35552c215ac78a892090f00be8"}
{"id": "train_038564", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d8919b61a630d249a5e805d7d5b59cb05d9f3a7c"}
{"id": "train_038565", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "69c5d583dc60c95a8fe53fc4d83c7d4259fe9c74"}
{"id": "train_038566", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4bcbf40c8af6797089dd81f4df4bafe2382020b1"}
{"id": "train_038567", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "dc20d822792c9fe4eb7a0ea4ef6e68362c8aea19"}
{"id": "train_038568", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["research", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "b064c112f804d38c91ffed6b2902d9ff39b52ffc"}
{"id": "train_038569", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "bb92290ca82a6758551c470365f74ff17771d818"}
{"id": "train_038570", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "4666dae49c5f2868a0c872312b96419b673491fa"}
{"id": "train_038571", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "3ada465fb3a0070975eae4417fb4efde3d7d07cf"}
{"id": "train_038572", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "b07c397dac5f77a711e915c58430dc7d02cc42f4"}
{"id": "train_038573", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "80629acbaf9ae6d87e6007805d88bf91b777daf5"}
{"id": "train_038574", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "9395303bb401e1fa151807da48476e2e177209eb"}
{"id": "train_038575", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4c07dcf8c0b7bafd4bd6ee22daf437fb4d1bd44a"}
{"id": "train_038576", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c9b85f107a8e65dfdcd9c5eb452e455938182feb"}
{"id": "train_038577", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "42fea324ccea39febeee5f4ed9c510f134694e3c"}
{"id": "train_038578", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "06e516c4c979b34fba472cd9987e07468df8bfc2"}
{"id": "train_038579", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a9578a6f1c6a2ae22cede2490a7a29fcb69e2653"}
{"id": "train_038580", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "0755758dc5b8a556dea54e246fea74440fa34176"}
{"id": "train_038581", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2c9e084d04ed7f9cabc27703dc43b4a6a1cb4fed"}
{"id": "train_038582", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "70b0945c8a9af31f06326c6fa488cba399f9b51e"}
{"id": "train_038583", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "a48faff4296c3872e45e930279d4c51b03297ac7"}
{"id": "train_038584", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "94f053f104d9ea4bcb218443da1f153359e44b94"}
{"id": "train_038585", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "e0aa878ba19366574338e14e70cc25e9eb9ab613"}
{"id": "train_038586", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "fbe02c19deb701725bb8b3060e8dd6353654d2e2"}
{"id": "train_038587", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "878b6a9398f02411b62db449ca85ec157fd4b0a0"}
{"id": "train_038588", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c9337bf7294bea9c6cd215c4bd1463148c5bd25e"}
{"id": "train_038589", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d4810073faa35e076edf4b757b60610f403f98b4"}
{"id": "train_038590", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b391c651ec9d5e9580d454005a6cc3fad2731a7a"}
{"id": "train_038591", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f27bc5c9e1af18bb1464c02012ba88a7919e1aaf"}
{"id": "train_038592", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["agent", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "36d7d03301a206b589492f4068a845b10043306a"}
{"id": "train_038593", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "46465a742a79c4b57c9d9d5b6d7c2039de975a23"}
{"id": "train_038594", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "42a16cdc3ef0774fcc7b257098bae8ab36bf1f80"}
{"id": "train_038595", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "51b61e698ca85b8d2190c33cdca2b1363ee9f153"}
{"id": "train_038596", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "c799f107e10cc190fa1696f52b1174ff415245fd"}
{"id": "train_038597", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "39f3795c19bbba594395a857c93c8b3e60950fff"}
{"id": "train_038598", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "db998cbf63f233158924add6e9fc00aad233a801"}
{"id": "train_038599", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7c57d0cea44d13f26fc9eaddeafd16360e4db228"}
{"id": "train_038600", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "77a4c791b629cb23ee06a6227b23a1bb533bcd7f"}
{"id": "train_038601", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "adb1f125b189a522cd619a0976db9872bdfe1c4c"}
{"id": "train_038602", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "4beada11c2131682bf2e0e32853b4f5e8eea8a80"}
{"id": "train_038603", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "bfd11c3aa3cb75dbc08f42be6a5dcde53e0651d4"}
{"id": "train_038604", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "bbc623f5df019221e2ffc1d6e78fbe58f844652e"}
{"id": "train_038605", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "1689a552b314ac86bb72b5c7b2bfba1fc0c7ac89"}
{"id": "train_038606", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "416c4f955bcc9f9f2fb64bfc17c5b69143fa2339"}
{"id": "train_038607", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f30117639fea108fb12533326dd8a8ec6be92215"}
{"id": "train_038608", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "3977b2ea1944cdbcf142b02288064cbe2faa8af4"}
{"id": "train_038609", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "ec9412579942a798bc8985c5ff9da0c5488b5c16"}
{"id": "train_038610", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "4a48f95524ade2a2e0ead54d4b5eea0d9d2c65ee"}
{"id": "train_038611", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "680eb2e6b1042209d16fde07a52c8cb46c45801f"}
{"id": "train_038612", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a3511847f6e5a6c18e82aa1774c69de60a683900"}
{"id": "train_038613", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "33b787a933f1346ad72ed21fa40d20464d8d4358"}
{"id": "train_038614", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a57c6d004f04f4a0c7bddb67cf5221221daa1fab"}
{"id": "train_038615", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "be70fe6db1686d69c026dc14fe33b229733fd51b"}
{"id": "train_038616", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e04243f825474e5a3792a36da3a600ab0bc67c70"}
{"id": "train_038617", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "4e0e1fd5cf9419f6d2abdf5efd1a707f9a92bc2d"}
{"id": "train_038618", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "3b8b7ddeec7032e7aaf592da871ce1d29ea1b037"}
{"id": "train_038619", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d2f774e8c3d69caf9003e44e3be0fc9d11bd91bd"}
{"id": "train_038620", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "c38ff50a55bcaef29cbe1d12eebb58ae477f5f1c"}
{"id": "train_038621", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "5572610aa236981ee1a59a2f31bea94fa444aa4a"}
{"id": "train_038622", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "ee798dc27573a0bad05badfb726a55eda27430e3"}
{"id": "train_038623", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "e497297075a0c30b11a8e242d16942fb265b578c"}
{"id": "train_038624", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d4cb72f379df6519d5a6030bd02f2a7d716aeb91"}
{"id": "train_038625", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "3b2aa9687a995a86e23ce16e142e84d3403f9163"}
{"id": "train_038626", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "887fcde6e63be1e969c9e0467fe29436b1f685a1"}
{"id": "train_038627", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "974c1db6eeb2e9386dee94b780ae18837af1d69e"}
{"id": "train_038628", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "6add7d9277d50b4836a76d398264545546c599a9"}
{"id": "train_038629", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "2709160c90c8e21ca59701de193fcfbcb37ffb11"}
{"id": "train_038630", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "2d6dd54d9cae93ed34dc29b4c2760e7ef915a0f9"}
{"id": "train_038631", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "4dd0109fe6d2940e1f03d29af2ebb0ac2b04e999"}
{"id": "train_038632", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ef62059cecc5e08a20e657a36dddf24a7e446fbf"}
{"id": "train_038633", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "1f78697c18566e12753469554cb542e747a7e674"}
{"id": "train_038634", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "8be4de57c505b41e69e1e7e06bf7c51725d1668d"}
{"id": "train_038635", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "92c3997aa33603c9af22e994668eac65fd6f6930"}
{"id": "train_038636", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "115f6e2d775d940cd51d27c4dc37e55b71257464"}
{"id": "train_038637", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c740ed73a689c3d127e65f6e87050ea1d1d5cfe0"}
{"id": "train_038638", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d511526b60ef8dcb5b3260516bfbb7d381cf10e6"}
{"id": "train_038639", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "a5bd3971c3df6c94a3f8653c303257d66253fdcd"}
{"id": "train_038640", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "354fafb4890c3cb2a5a413fe5c9c9ecbe8b32be9"}
{"id": "train_038641", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3ac6b3f17e6eb65309665bee6fcac11ce7dfb586"}
{"id": "train_038642", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a95e66e74197e52f3e9316bea09cbd3207f7c18b"}
{"id": "train_038643", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "d3e2b46255c27a92e0443ec63df864a84001b97d"}
{"id": "train_038644", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "556814808fbec9ffe0ce95a2866a6b4a43a5b5c1"}
{"id": "train_038645", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "6ff168e7ec6e1345e1af256d647fa3aea9be7c57"}
{"id": "train_038646", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a46ef1b49bc587e228c1121b2fea73bee2f5c156"}
{"id": "train_038647", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "b0920636e701b9f04a7fe50b49051bcfe8ef3364"}
{"id": "train_038648", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "aafcd12689ff6f0b887bf3970f24e4be7007488c"}
{"id": "train_038649", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f165a8a0099a6be3242f88305dffb16d67b80c79"}
{"id": "train_038650", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "9d9de1fa32dbbf7387e9bef523e702ca8541ba22"}
{"id": "train_038651", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b84775a95e54a0c896a63a4a20bac9970fc8e7b0"}
{"id": "train_038652", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4eb1063cd49d27ddcd8620aa8cc19eba3c560a46"}
{"id": "train_038653", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "90169fa78a4a64043829b2c533fc0f94aceaa354"}
{"id": "train_038654", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "c1dd27bbc58e5f7160b431dcc6035316f6b7a6ed"}
{"id": "train_038655", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "af16ba7d44f512e50e7adead691785ead5247bf6"}
{"id": "train_038656", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2c40bf61ee35834cc8c94a399da7390d2b8cc904"}
{"id": "train_038657", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "6011bf2d8a331baf555d63b68b7ebfc38d8a7bd0"}
{"id": "train_038658", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "bdc384b230090c50841f7b05aae667bb85e138f1"}
{"id": "train_038659", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "df0a2f4fdbf621bbb6268d0ed57249969278c872"}
{"id": "train_038660", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "ff761a35db98902f6eb8f58809a920b9cf3926de"}
{"id": "train_038661", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "768b11cdba16487b75ab86f95f6cf2b0e03e726f"}
{"id": "train_038662", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "cc043ae63e8933e4e7a84920aa97424639314748"}
{"id": "train_038663", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "322b7e4f8b8d86e9a15368c39bb260243a6ba0db"}
{"id": "train_038664", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d8623bd4303c33853db78c2482be59effe9f255a"}
{"id": "train_038665", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c2427bf736afeb4f98eb1653b6c2f43a785ff2f9"}
{"id": "train_038666", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "31506cb1d0f68317059ce232557d0a6718a8ed7e"}
{"id": "train_038667", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "27bd0ee398566ce6a624d13a7127aa230c727a8c"}
{"id": "train_038668", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0b823a34f41743ef0a494798650135c5693ca92c"}
{"id": "train_038669", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "1e056b810dce42cd27b4aa64380307ec9ff2960d"}
{"id": "train_038670", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "135804d415b41200a7a862bf6f40c612485a8dd7"}
{"id": "train_038671", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "63234ecdf3fb430f52d5d342764947f6a31a9012"}
{"id": "train_038672", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f84f5877b0c170758aabaef93476427b743a9706"}
{"id": "train_038673", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "70d78986822c7500b779b5d520adf99efda79cdb"}
{"id": "train_038674", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0700dce5015ef3f8f9b48604554ead2e16c9a4de"}
{"id": "train_038675", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c6604c18c2b782e1486bd3d801dcad5303822774"}
{"id": "train_038676", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "2b3fc26b3428845c6aead6b3cfe059f079caadd6"}
{"id": "train_038677", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "a871ca5cb56a66f63fed7525265e5de60ee9b6c9"}
{"id": "train_038678", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "84bd4388c2f92075eb488c704e0b8a2008d61463"}
{"id": "train_038679", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0b656701845248d518cbfe8f1a14d6f908637b5e"}
{"id": "train_038680", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d9e031d07ff0102d0acc16bb403ec714378a7b70"}
{"id": "train_038681", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "abb0ec6e65f5985879b6208d86d14c1f7834b2b2"}
{"id": "train_038682", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c12cf3d3e069fe3d4a7295a6faba1d9df712f61f"}
{"id": "train_038683", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3907e057f69d47874b47e92f64c06a3787262f9b"}
{"id": "train_038684", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5b91297cd84ec7ecd71456ace2c326785f8f8f42"}
{"id": "train_038685", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "289a7568d408fece2b28cee89fc7407a2155eda8"}
{"id": "train_038686", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "eb16fcba00b48bef219c287f8e9891983d219214"}
{"id": "train_038687", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "69b04c417f2917376431d8d54f4dec28ee727415"}
{"id": "train_038688", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d8919b61a630d249a5e805d7d5b59cb05d9f3a7c"}
{"id": "train_038689", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "63b84c08317b79900b1edda4de2fc92122b7af74"}
{"id": "train_038690", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "28b6c4aab232933797a96e08995b2e0e6f175766"}
{"id": "train_038691", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a9dc23a13722d16a53b816c3f734045dc9368b57"}
{"id": "train_038692", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a2ff5d8cd646db9b690724acf0914a35aa5275c1"}
{"id": "train_038693", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "899f85fee3e1c2f99953666db851f644a5730b38"}
{"id": "train_038694", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "4b4204c89610b8b51f6b8ab5ad0239f76134d010"}
{"id": "train_038695", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "2791f665838fae460710e97cc0fd6afb2974360f"}
{"id": "train_038696", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "3d849c623b1b1d5fee5c54ee6c8d86cb44cc1f28"}
{"id": "train_038697", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "5b4a5e4aa8e8126369ad3cbdf88f40f9c8accb4a"}
{"id": "train_038698", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c4869bca68f9bc4c03a03c0415a17f49368a718f"}
{"id": "train_038699", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "5e46b4e82ab72279bc625cf5d5a4c8e31c29a081"}
{"id": "train_038700", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c2e04b675140fa68f4e11486a89ee66f8d8ee833"}
{"id": "train_038701", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "b69df4556637e279e86b74df2a736fc7e7c77e09"}
{"id": "train_038702", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "554787c2558320679f6d042ed450e985a9332d7d"}
{"id": "train_038703", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "4039bbf36da67fe7901eeb1095072bb39792ff83"}
{"id": "train_038704", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a82891410de390ea4c587d2e9373fde73f132e3f"}
{"id": "train_038705", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "af221d5529e66af03d2a4383d0891cb920c949b2"}
{"id": "train_038706", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "754454045f575eed22d185938715d56c1df1fb66"}
{"id": "train_038707", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "379d385156d82ef49b6666c810dc27209aba4322"}
{"id": "train_038708", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "c54b1c14a03b4c0d46c278d02eb814c6f4e02cc6"}
{"id": "train_038709", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "0a04f134bc99df20ae1517a4efe46c9f0e991950"}
{"id": "train_038710", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "1c0be31b09a5d77e88f29d286ff7e900268ec7df"}
{"id": "train_038711", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "fc334d52f193bac18c35da761c59165ee2af6693"}
{"id": "train_038712", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "272965b56b30d1605b11013ebc2aa77df678b2a4"}
{"id": "train_038713", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "c07ab4eaaef77de49ccaf92f69746b8c977c9aed"}
{"id": "train_038714", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d99c092be5d8f78ff99fae2d748a5f98a6a51b6c"}
{"id": "train_038715", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "625342666a44f944adc361cda9a2ff80ddb063c9"}
{"id": "train_038716", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f53368a0032469ded99bd99b85ec80a0e022cda5"}
{"id": "train_038717", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d99b4b654a86fe1212c3af3f1da37c29beb8efbf"}
{"id": "train_038718", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c686e9b148bf580f22850f3540b84fe71b25fe95"}
{"id": "train_038719", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3acaab01339372c76fb87c7b243c9c491ea159ce"}
{"id": "train_038720", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "33f55b1e963c0e3a78b42ab0a757e0d0eca9ca61"}
{"id": "train_038721", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "dadbf0559db185149b1478ef334d5999dea2cba7"}
{"id": "train_038722", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "ee8372cb9bf108ac99520c29ac6259b0d5e6533f"}
{"id": "train_038723", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "5bf199aa5b3e635a63995e732c044aebd394a1c5"}
{"id": "train_038724", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "beb22ba1130b16542e18cdd6116db4b3f20c2bac"}
{"id": "train_038725", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "40ecf82151e7a584f149d8b385f8275093e79e9c"}
{"id": "train_038726", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c96e0a3e44932fcc1048d213911a8ec2dfe9ab3f"}
{"id": "train_038727", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "8720dd8b0ed3ce5f2fc52f15e3d32ec1ba057140"}
{"id": "train_038728", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "01244c0430784b2ecf3080770b1019e2d3fe47e0"}
{"id": "train_038729", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "35db3d223e108e90fd6f8380e6a2a19319e96c4a"}
{"id": "train_038730", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c67c143e3e0eda7281d4888ef85b2f2b31159b83"}
{"id": "train_038731", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "530a0b523b2e37014e017b43f8b9f0d630af26dc"}
{"id": "train_038732", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "0a7433e4b5789ebdcd9a684538706890fe639816"}
{"id": "train_038733", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "21e4622704243820bf04bef59730b6d75ffd4a73"}
{"id": "train_038734", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b0791045ee1043ca69d9be0b45234081e453322a"}
{"id": "train_038735", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "62848eb04203b9cfb926d5a35d81e1acbdd31465"}
{"id": "train_038736", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "da8cfb30ad82b3d5d51b2a71c6fcc2daf25d3d46"}
{"id": "train_038737", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6ed063afb80d0da179d974c628701f67b2f91836"}
{"id": "train_038738", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "1c564cb6aa5627d5000c06df42b09dee548b4d85"}
{"id": "train_038739", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "c19189f159581c0901b85b72cf232ab480dd26de"}
{"id": "train_038740", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "ec6cf59ae9f612a2763d1296b0418f57d5cca4fa"}
{"id": "train_038741", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "20c132e3beb119b1d2ab68039f03c8d3cd97a862"}
{"id": "train_038742", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "76c4a499175b29c9a08d237033bfac7484bee6ad"}
{"id": "train_038743", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5f2ad6c8e5ffd0bc0c06a89cd76ba03ecefd5899"}
{"id": "train_038744", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "f0e2ddc2edd7ba9cacff7acb9e670225cbe9a94c"}
{"id": "train_038745", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7ffa864f855b58aea74abf018519244f460fb506"}
{"id": "train_038746", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e7a5e3fd8509ae6829fbcdb96152e58bbf0b8ea1"}
{"id": "train_038747", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "42a935a96c03f4d649d6fb7d0623a9f1d3f5b580"}
{"id": "train_038748", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "5627330a3d3f5cfb1a2b39e9c64602012d19d019"}
{"id": "train_038749", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c4c67e811942317b762b995d80e790ad55711ebf"}
{"id": "train_038750", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "9eb1a449a8f0bf77f47c53fe939572a89ab9e6f5"}
{"id": "train_038751", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "14f4f6b84fdeb73222ad7bae75a464d78135040f"}
{"id": "train_038752", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6f2cf590cfcd313552a7612fae4c67a359ad2ce7"}
{"id": "train_038753", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2b18f704ac63c4005dd7c49a6629f13322847e63"}
{"id": "train_038754", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "b5c19ab57afe9a074c4c773c6bdd267ed5f96811"}
{"id": "train_038755", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "9fa8d32500e1a8d5d12cfe21a5ece5b17d2624fa"}
{"id": "train_038756", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "63ca3536701c444a6eafaeb4439f30c7de58fc9d"}
{"id": "train_038757", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "616085785a0d15d86890b50429d554629e2ba1e6"}
{"id": "train_038758", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "50aa57d1287460ff41bec4ed9ad04252018ca431"}
{"id": "train_038759", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "0f76576419b75bfa7dc802b2f213b3e1b541d677"}
{"id": "train_038760", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e565d5793a4a21c459bfed9d353b035e7a3201e3"}
{"id": "train_038761", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "3c11277c64d5e34033d5cee4e9424da19158cbe5"}
{"id": "train_038762", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "7c71863651524d8f5a9a89af851e5b6b05728a2a"}
{"id": "train_038763", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6bfd73ca33a7ca9cd487096b7a51fa0099a3d882"}
{"id": "train_038764", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "8e58b4e3a3086e9e99a12d0bbb73b15897034c90"}
{"id": "train_038765", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "76ada0ff613cc1b8e3222a81b118789ec4035a91"}
{"id": "train_038766", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "160681b7fece773c1aac141625e90f78b035cf58"}
{"id": "train_038767", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "fdfda76b521f2828ecea955eb46eed36e759a178"}
{"id": "train_038768", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "590d1294b42d4bac271a0dadd73505700c842802"}
{"id": "train_038769", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "039edb289f6fc3622228c6f084e0dd8c0db58296"}
{"id": "train_038770", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3d8585f68f3b8406ebb13c49adfda77a77d62859"}
{"id": "train_038771", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "819e4932fca50f69c8bfa3cb0d03438cfb45fc07"}
{"id": "train_038772", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c7c8fb7f1e24bdff649a7f543969109dd7f0e12f"}
{"id": "train_038773", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "92a69375df9aa8c64348537491a576dcd36a36dd"}
{"id": "train_038774", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "4628723b326eca9be0514897e6f6be0013d02b9b"}
{"id": "train_038775", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "20872573475dc6041dea1cb49636fda9bae4dcb4"}
{"id": "train_038776", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a842f5d8b003f24b28203bdef5cbec7ffe0d0b12"}
{"id": "train_038777", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2325faf740603a82df066d181fb1a314308c662e"}
{"id": "train_038778", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7fdc534b4da63b78a2a44195aefa3ee0cf76a1d9"}
{"id": "train_038779", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "d39ceaa4ca505cf43b29e68429b2e2d8c8619275"}
{"id": "train_038780", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "1337bbf8497c272d7cc6c86319d03fef67dd83d0"}
{"id": "train_038781", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6c4e353e6f8e469c552dd2190d9d710b44a3ab9c"}
{"id": "train_038782", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f7c1cc0891f2abeffb633ed6ab7cc43df9025bc1"}
{"id": "train_038783", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "7b4f4a94580028d2e2120365e449dbe66178a045"}
{"id": "train_038784", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a371a084572bd791956c59e0eb57b3b7c07d63e5"}
{"id": "train_038785", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "58595150a48407b5f96fd53963fe761ad9e217f4"}
{"id": "train_038786", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "04d7c488eb0413115ca82cc3efc23380c70caaa8"}
{"id": "train_038787", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e2560db4f15202bb0b0cc6e7e15fe59661c98617"}
{"id": "train_038788", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "fe38c1639a08e32893002f34f6f913f3f0327b32"}
{"id": "train_038789", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "81f3a20ddaedcce4fabd1430059844be75b8e102"}
{"id": "train_038790", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "513c6f062141cdea5b96b59821bc029f63862886"}
{"id": "train_038791", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "15e689c3d3d8672132739ded0e40130e1bb19d55"}
{"id": "train_038792", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "79bd784ed9cb863781889d7b56263f7de1607fb1"}
{"id": "train_038793", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "c9c7e4ea8d223fc373a8b6bd8f542578111bc465"}
{"id": "train_038794", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "d00d5c765058b7ad0cd7a7941401b246d1d71d51"}
{"id": "train_038795", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "83dea4ab8f0cf3e99cc216d41c9aea1d76d3c670"}
{"id": "train_038796", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "61d8e4f042e6b467bb6249a155d9d3017919e6cf"}
{"id": "train_038797", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "b9e277a42531a4b06684c39e3e2d8d9e6a189fbd"}
{"id": "train_038798", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "21e822a54ce67cf72d0a75703b32a13eb3cf4912"}
{"id": "train_038799", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "759177d8ffbcd49894bd320877f61bb542ff75c0"}
{"id": "train_038800", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ca8a7a001b470c2ee4041fa628b99b431689f723"}
{"id": "train_038801", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "28fb5c30e3709c44ba302fd4ea6ce3247d883bb2"}
{"id": "train_038802", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "954631db8d44ac7938d77fb9d8248c804ef478eb"}
{"id": "train_038803", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "16871cef1acf38ace4f9e0a1e304ce0fff5528cd"}
{"id": "train_038804", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "d6261ae344b31591e7f2b8c6410068239e3c2cfa"}
{"id": "train_038805", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5e46b4e82ab72279bc625cf5d5a4c8e31c29a081"}
{"id": "train_038806", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "c2e17d1eaaa67204f398d73e91dfa95b526b62d8"}
{"id": "train_038807", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "1bdc37d4b47d89a78f2ade1e858524c3dc6846a0"}
{"id": "train_038808", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "fa996613ca411e20eaeb2ddd4a8d141791eafe80"}
{"id": "train_038809", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "e3798ffff1addacbf4d969261f8d1393b3a73cb0"}
{"id": "train_038810", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "64f8ca3cc9cc3ba31aafc5be0119ba082d274805"}
{"id": "train_038811", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2b99750dac57e46daf7d149c8c33ca443212130a"}
{"id": "train_038812", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "fd44524e55bdf40d1a3611781868bd554c9c1289"}
{"id": "train_038813", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "dd65d9f9d29cbacda249a84b78a6e32a65e87d2b"}
{"id": "train_038814", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "896528dc1333d278e3c196e444dece6acccbc2d0"}
{"id": "train_038815", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "5f43a2d864b3109a5bd7fee88596227d201478e9"}
{"id": "train_038816", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "75f205a378bfdf3c2d2177e4fd56fd74476c0e46"}
{"id": "train_038817", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["agent", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "d0f9c29e2258ffb66c74db71129c6810b37cf5a6"}
{"id": "train_038818", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "1be1c8cf54e58c051e7e74f64b8b3bfdf38ca106"}
{"id": "train_038819", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e8be2768f29d76ec9e3d95483155a5e6ecfa7541"}
{"id": "train_038820", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "b056c3de31a55e804e32a9ef1d62c966f64fcba1"}
{"id": "train_038821", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "50f1f7afe13d8978c3ef3c674394287c41f1fd45"}
{"id": "train_038822", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "42a16cdc3ef0774fcc7b257098bae8ab36bf1f80"}
{"id": "train_038823", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "fb14b8ca586968424a38d90afc74ce5bf74d11e4"}
{"id": "train_038824", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "6d12a7e25c124cfef200af7e53966380100631b4"}
{"id": "train_038825", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "589e42786c46723c9010da65713c8bab0a6d6ba0"}
{"id": "train_038826", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ff4bab78e80c9a9d9bd830443d2b123386cce057"}
{"id": "train_038827", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "eadda9dd521296e496a495863cec2f033154a601"}
{"id": "train_038828", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "1eedfadb1b3afc2d49136574bd522c5a77e9dc6e"}
{"id": "train_038829", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "54f1ebd2b723ab225bf52fdc8114a06313a28cdf"}
{"id": "train_038830", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "5da5fab315994ea277695a537dde33442bae2bb6"}
{"id": "train_038831", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "4169c8f23535e35bbd9a5960a51abc0be8493aab"}
{"id": "train_038832", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "659ea3b9fefdd0f519edafe434370c5dacd1af5e"}
{"id": "train_038833", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "9540ce2072c721e7dc70d49ac79a4b9cd7475d0b"}
{"id": "train_038834", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "226b87d2c56d02dbef25e9349f49a880df152f91"}
{"id": "train_038835", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "580f5fbce6ad69cca3af2af9c8a91941adb12d06"}
{"id": "train_038836", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "410ddfaedf7a9b2448631791f089b54f456681b4"}
{"id": "train_038837", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "020899d20a27a2fa5651ba40a50b9ac4bb6052ef"}
{"id": "train_038838", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "08e9297aa9f4690b4e808e456fa1954d247f2349"}
{"id": "train_038839", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "40a757318a1daa41a9960c34c080ad73a89982f7"}
{"id": "train_038840", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "70dbfb81a07a9bb8cdb956f182bafa17a0765b62"}
{"id": "train_038841", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e46d9b5398526b6ac63185f672ff8e348391636c"}
{"id": "train_038842", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "1aab72284b4fee0e9b821117d790bd365ec0c683"}
{"id": "train_038843", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "1ca2f2c261168249668875f139b278fd241f7f5f"}
{"id": "train_038844", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "12307de06c69fb439d7a2faf9c41dd6f8bf37cd7"}
{"id": "train_038845", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "62949c7c15652a54d636060dab82da713458fc84"}
{"id": "train_038846", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "2f0425f1bd64ab2dbdb3d0364eafec8a59a1f223"}
{"id": "train_038847", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "1b83efa60291b5ba34f34fbea1c55777d71f6106"}
{"id": "train_038848", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "969d832e43a71ec58e885144ff560db45ee5fafb"}
{"id": "train_038849", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "0b248d66f7d8b64fec02cf2913e8a8a269d4a3d6"}
{"id": "train_038850", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d34566264bd00bc4f99d9b559fbd311c33256892"}
{"id": "train_038851", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "c6e577fee087846667e885d79419d117fae458b7"}
{"id": "train_038852", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "9d41b8019f3e70722f9201eee5672ca937413fe5"}
{"id": "train_038853", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "9d360739345f8d5a8cedfea6a10c388b543acb6b"}
{"id": "train_038854", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0dbdfb7219bda6e40a977853b3739d6149a34a8b"}
{"id": "train_038855", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "5bfca95102c26c9aaeb8e38d5d12246031412cb0"}
{"id": "train_038856", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "ba64c04beee82f63d89fa75c33a908f402ad4c17"}
{"id": "train_038857", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "eec5ab7782073cc10ba0124890d9948b123de8f4"}
{"id": "train_038858", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "e09cf3d7061d5b9743de321c1ec647779ef001d5"}
{"id": "train_038859", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "82313282548f199e875aafa581297d1d4ed3cf51"}
{"id": "train_038860", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "86056c4caa7412abb7246293b985975235be18cf"}
{"id": "train_038861", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "2e18b6e481b91d07f584b54cacaaf281179c1743"}
{"id": "train_038862", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "073165894f90fac274fd41124a23db323389bad3"}
{"id": "train_038863", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "4f321550d2872fc933d3ab0cca410f82e64fbabd"}
{"id": "train_038864", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "ae05b6413aa7f8330a58699b6d341b2caa622523"}
{"id": "train_038865", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "01da951f9a1dd085219b01b210f4916df612f15f"}
{"id": "train_038866", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "95909314a78b465ef88bac5698ef2c5ce4285e75"}
{"id": "train_038867", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "aca474e327f3b00a1e0902591b31a6ae3c7c07e9"}
{"id": "train_038868", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8d3abb684cf92f7354c4e3eace4abb735e5b1030"}
{"id": "train_038869", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "040742ddaf7350200696595c63d804214a5c07da"}
{"id": "train_038870", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "893498c0a1623170d740b5a63c4726bdb47b22d4"}
{"id": "train_038871", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a83fc5c9139876b0b6214ee215d2421e8a75bc0f"}
{"id": "train_038872", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "63862e15d340b3417705d7f6af8d5ee762fbf368"}
{"id": "train_038873", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "324527ceab358ec5ca7f59d71ee5be87b3c81739"}
{"id": "train_038874", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "21650e0e9a89e6eb40fdbecaf9c7f88de262e087"}
{"id": "train_038875", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "0d864ea8aab9b9eaec332aec0c163c655939b4c9"}
{"id": "train_038876", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "85fc135bad5b80fd23749ff292221944d4350012"}
{"id": "train_038877", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "913262744759ae110d31b2c8b7f759ab5a66ebec"}
{"id": "train_038878", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "40905b1c088c3de9ee2ffd118ceabc9c768e016d"}
{"id": "train_038879", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "a26d367c085e0c10a7cfe4dcf015282567b1ee21"}
{"id": "train_038880", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "47ade410926fe60a6029ffe7764696edd23d6bcb"}
{"id": "train_038881", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "3296aaa34798b0b9c0f0600ef84c913cb824bf5b"}
{"id": "train_038882", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b7ef91aa1ca9591cdea78e42c6703d2971adc8ca"}
{"id": "train_038883", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "7acbd9292c3663db247ad9e55047bfcd3ecf3fa1"}
{"id": "train_038884", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "3118d4563dc6523ce7e07ba78a529ae99b643cfc"}
{"id": "train_038885", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3253cf2eb112429f3fc76c39d56c966cbbd32d04"}
{"id": "train_038886", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "47240b16ada1a22c47ea8a09406a5f8f646ae670"}
{"id": "train_038887", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "41f69f4bb7f1a74f719efa5ec6e33725115398ce"}
{"id": "train_038888", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ed8c88f406e9855d9bff81fafe19b61f7cdf8bce"}
{"id": "train_038889", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d4b6cfbd6e21326e7475bc8471208ba14f98e89c"}
{"id": "train_038890", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "71d656991defc3555435903d9dc29deee2b66e63"}
{"id": "train_038891", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c41d9eb8b0cb86d2f713bd665372d8703ae43e9f"}
{"id": "train_038892", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d9a6bb982d0d6bc7508a7d7861d5926ef080e97b"}
{"id": "train_038893", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["performance", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "3b214d8c3d0fab7a440ee93fbf8291853135e7d0"}
{"id": "train_038894", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c04a274bec4d1eea6a73755e99c496c43f2893c3"}
{"id": "train_038895", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "0ee66f5b611a5db3b930b93e990b312ac8f7e6f1"}
{"id": "train_038896", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "dedb70fe20d68327caf36d505145a4b00966f81a"}
{"id": "train_038897", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "89e92c9f3f4a0d0f99b25edd53b298701ba158ad"}
{"id": "train_038898", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "91aa7fe284bfd222eb9392398b42769c36592f9b"}
{"id": "train_038899", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "fe668e7039aeb117a35a61995fe9a758ac98b534"}
{"id": "train_038900", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "bd940a6a7380c504ce2b479763568a6f3975693c"}
{"id": "train_038901", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "60dfb181e06eca47679a5c0f455dc9bec90fa25a"}
{"id": "train_038902", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "e636044e5f1bcbfb03f4616a8641bdb8a71ed2e3"}
{"id": "train_038903", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "16368252d4a828d8853b04b31707a67071dd167d"}
{"id": "train_038904", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "656c72f11d3b452d9c683fb5f8421354560bfdbc"}
{"id": "train_038905", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "4681bb93f966b7c7624c84bdb08b662cbb0564bf"}
{"id": "train_038906", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "56174617c7292e0dff3c8ac45386c1b275f3bf3a"}
{"id": "train_038907", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "65c534be723aa72a7a7cc0f241a12a0e7909ae14"}
{"id": "train_038908", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d9019e54144d513edf30a3668c969d8a0742d20e"}
{"id": "train_038909", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "4a9b0723ceff247c443ac2f2b70535bc6a1f5e29"}
{"id": "train_038910", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "e23252377cd12e1bca2f8c9b31aa902d5ddbe4d7"}
{"id": "train_038911", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "3c2affab7449849157261abfa1ebcfd2e80f4ac7"}
{"id": "train_038912", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ee79960df4d151c6567b477737ee77961ae4a45a"}
{"id": "train_038913", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "44d5af9d042cc33252724e5569a2043789908a76"}
{"id": "train_038914", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6665d28c39ca319aea392b24f7bd54d149ee2f4a"}
{"id": "train_038915", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "54302977e6f2ddfd51537a45e11964fe71d97743"}
{"id": "train_038916", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "fe65473b08d70009fbc6508ed3fc6f036926496a"}
{"id": "train_038917", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "b4a3e118d3cca77c2ab58d43333c6e93ce2607fa"}
{"id": "train_038918", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ac7b4c4f4b4acac8fe2e55318276c18ea0e28a38"}
{"id": "train_038919", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "b770cb02c441fc6bc45ddc8cb89950d7dde745ea"}
{"id": "train_038920", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "164639556e0ef00b032ff763cd553395896fb91e"}
{"id": "train_038921", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "718a384b40825883b6e5e862de3ef6cdb9f1004e"}
{"id": "train_038922", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "87df2d8d84a8c40cee8f73e35c21b80cbc6b7738"}
{"id": "train_038923", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "df7384ba8d98d40b1c7bf24f9e5fdf331642ff5e"}
{"id": "train_038924", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "1c3e85a7d3d9f051393d4b51d5b23f73bf15e9ad"}
{"id": "train_038925", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6d17c50e66babffb8f9958013db2f2953477d52e"}
{"id": "train_038926", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "68b3c102ab32f92d19397bc671be39b49b7d6248"}
{"id": "train_038927", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "fa382806a9cbb61f5e3dd3af1c1bb3bd1c762f9d"}
{"id": "train_038928", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "3683c7705e6951a0d973b8afbf02d5ce71cd7b71"}
{"id": "train_038929", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "07cc95a522740cd9f7d4058ff0c12939eafe6065"}
{"id": "train_038930", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "7adb10256b75668b071b5b6fa554991b1a6ebe89"}
{"id": "train_038931", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "a57c6d004f04f4a0c7bddb67cf5221221daa1fab"}
{"id": "train_038932", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e5709846b822dce0d8c9c7649eab51c048324fc7"}
{"id": "train_038933", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a4da0ac24d79aa9cba1d3836bfc26d81b516b35b"}
{"id": "train_038934", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "30559205ff1171096a4aff0e98ce1433aa1d20a0"}
{"id": "train_038935", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "010cdb5863f4442e66c953c8cf42e7447ea5b898"}
{"id": "train_038936", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "3d3b097eb5e6fdd5cd5ab8d9f91e70c88cbd1e3c"}
{"id": "train_038937", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f9a0404d6ccecd63d6a0304ea2c02b0173f57911"}
{"id": "train_038938", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "08eef8dedeb2e60b7bbea5005af36693e5f55a38"}
{"id": "train_038939", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "757e8958969b7ad0dfda593498fbf3f9cad57ebf"}
{"id": "train_038940", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "53a71df6609de6937664b5badc20d1990fc913d3"}
{"id": "train_038941", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d0c9dc7ee15bff60d6b822501aa34d0f6b789ab1"}
{"id": "train_038942", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "9463465f3cc5cb478702f457a821e664f32f7876"}
{"id": "train_038943", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "d84ef6aff6ae9118f027e614d7f5ead7f8aaaa19"}
{"id": "train_038944", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "d0f2f79aca34383611046a65189e09a760f16dc4"}
{"id": "train_038945", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["performance", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d0e00ee8dc47ee16ed2a47111b4a2fb5c4479a73"}
{"id": "train_038946", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "ffc4a06114ebce85641cdcb46b35d7be010fe721"}
{"id": "train_038947", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["research", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ee2b803f58f8fdff0deeca7b53de30443b650517"}
{"id": "train_038948", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "3e0b4cca56013a90214632a4efafeb79b9da2983"}
{"id": "train_038949", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b4f98b0c54a6b3935d3ddadd05666a4a42fb6ac2"}
{"id": "train_038950", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "855bfaafef825793152389b39280e11a56ae1a80"}
{"id": "train_038951", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "18526cd2d5ab426f815ee48f4f40d36bfa66f223"}
{"id": "train_038952", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "30dadf7ed58c5abd77b5dda27d9d02cdc1f43b4d"}
{"id": "train_038953", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2648f0393fc44a767fccd23239ef96f403ff9297"}
{"id": "train_038954", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c0ab191a0027f385a40852d791b55ce370f5f4b6"}
{"id": "train_038955", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "8bfbfe8f667ebc38514fe43bfee55bb52f044c38"}
{"id": "train_038956", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7938c50553247f033057e6cb5bbb82dbcac41e56"}
{"id": "train_038957", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "0a68e648d10fda6c25b303f77be7bad3a77e1c76"}
{"id": "train_038958", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "cfd684551a3eb3a640df128234912473af9162c8"}
{"id": "train_038959", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8e2b17f448d2763a048d5893b564487496c1c96e"}
{"id": "train_038960", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a44db2c6d3dbff7ac7d16a6e6751d2d5ba4bd322"}
{"id": "train_038961", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "4accf697455ea24bdba13542cfe6b30cfa552922"}
{"id": "train_038962", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "fbb831d1c9d9cece6ddb83923c6b7f690d489708"}
{"id": "train_038963", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "8e074a971e45358c67729cde288337386adda6e7"}
{"id": "train_038964", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ba2bee17a0025e6bc60024ee4963ceeb16affe30"}
{"id": "train_038965", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "09cb15711202218c2843e9bf623c904080c3369a"}
{"id": "train_038966", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b63d1faacd1a9c5bb0636bf8067db49a355916f0"}
{"id": "train_038967", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "df5abc138d8b72b39ab2b807c3c56c25a7b57ec2"}
{"id": "train_038968", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c10d1edcaefffe1b1e05b046018d0110d991d835"}
{"id": "train_038969", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "868108214c8a0522c326623dd074ec767278f9ac"}
{"id": "train_038970", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e9f8b9bce1422aef984d8bae9f3a1306ce202f8b"}
{"id": "train_038971", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "a9edfbeca8ee9dd9e87bb4c7872dd8182409244e"}
{"id": "train_038972", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "25e304abf5d25c9204059804843a295c8737969b"}
{"id": "train_038973", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ba554b1b1c30b5a087224f412aefee2dbd115bd8"}
{"id": "train_038974", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a855b60dea221e6352f2214fff0d0e4c6f60e45d"}
{"id": "train_038975", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "6d073ee1e6e395f3474cb6d94062f704aabcdb16"}
{"id": "train_038976", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "1eff519f967f72576d15f893486108503aa6fbd2"}
{"id": "train_038977", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "f658721074960b475f1542647e79fdc03d20b127"}
{"id": "train_038978", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "67581b9c2389174c49ce5d5f96510d705289aa34"}
{"id": "train_038979", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6243243945e8b1e2692bfb6c268b36c2e72c79b8"}
{"id": "train_038980", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a4c2f003a8601c4cbff460851e2edb6ffe372096"}
{"id": "train_038981", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f851f5319165c8b9c0d3b4504d6a41f89daea87b"}
{"id": "train_038982", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "2fbf5a34a9b31916688547f2a8563a2da50b4535"}
{"id": "train_038983", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "ef9450f1fcb77010e96aba3daa09c598f105b709"}
{"id": "train_038984", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "57de9c8205788c8369a5e0c40634421385a81fb9"}
{"id": "train_038985", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "5736657c44ece8b63bc2db6e066efb9b0ada1870"}
{"id": "train_038986", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "0413f19ee9e35265b697c7d2e6e622a5623db24e"}
{"id": "train_038987", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4c01cd635442bed9bcf8ed6bc407f6bb0802c100"}
{"id": "train_038988", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["performance", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "fe004da59b82425f36f5f7be4a379d72af7b69c5"}
{"id": "train_038989", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "fdca3cbf8cef3cf7d66fb5a96ce4e3115b97cd13"}
{"id": "train_038990", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5f55dc3f44aae9963d402720be14566348b36942"}
{"id": "train_038991", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c957f53621dbbc272ea991e2f1076235402949d2"}
{"id": "train_038992", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7a1271ab955a29f1ad5a5ca2dbeefe99ec751238"}
{"id": "train_038993", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "b7d1f578ff59694c698e20273140f2bf6e9db796"}
{"id": "train_038994", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "d5c0dc78f430053e33f4be40a4e39676070dcf60"}
{"id": "train_038995", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "03f409eaae2f2bd2e4933a7f07532e0253764afc"}
{"id": "train_038996", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d1a3b108fb1a68103d6625c69cd83f905f84c2ee"}
{"id": "train_038997", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4cda20be150e6157ee24e98c15fdc02844aaca08"}
{"id": "train_038998", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "53f22e47271dce66d7cc697705d1adfc8f1b90b8"}
{"id": "train_038999", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "1ca2f2c261168249668875f139b278fd241f7f5f"}
{"id": "train_039000", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d0e00ee8dc47ee16ed2a47111b4a2fb5c4479a73"}
{"id": "train_039001", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "cafe6cf5c68a69e661801cd9a257ba5e03f59e4b"}
{"id": "train_039002", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6d2da016e62b792f9e808bdec69b021935228cd3"}
{"id": "train_039003", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f7c566324f62f0cd6d1dd46e7d225b9b70b152a9"}
{"id": "train_039004", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "b8f1a53872206cb939408636b0e6a2e443aa1cca"}
{"id": "train_039005", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0821b4565a8050b5a6efe5a856ee08196c43d7f9"}
{"id": "train_039006", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "85361f59c8d351e71b4d137feda2b07648261732"}
{"id": "train_039007", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ae6d4a8fab7023c0fce0a97f8471f7ff6c3a24c6"}
{"id": "train_039008", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "7baad225c5da6ffb04244aa48a24098e31ed51af"}
{"id": "train_039009", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "052e73a210565c087ca374df4b4dc0bd72a2713b"}
{"id": "train_039010", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "28337c66936a434db78e078e15edd6bb764890b1"}
{"id": "train_039011", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ef8f8b0713a1a5fe0e739d83abb186d362f66cd6"}
{"id": "train_039012", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d3152d989e287ba33a5d97358215c03d84fe9a53"}
{"id": "train_039013", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "7e51ceeffb3571aff1cf8b337215a52de0e492a8"}
{"id": "train_039014", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2103ae3270b0675329b3f1b6271e7f107ead9099"}
{"id": "train_039015", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "26af262c46fe82c80a7e85a125d68726d3cdd427"}
{"id": "train_039016", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e8c59c7e5e824de0d9ea6255f971776f9e8c3c0c"}
{"id": "train_039017", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "12b21a8e9792ddfbbd31f3ac5d159f33727e0b8a"}
{"id": "train_039018", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "4e48141159233138b00238476875b2f86d921dcd"}
{"id": "train_039019", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d3b475afc89710a62ed6beef3a796994904015a4"}
{"id": "train_039020", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "8e3c3a6f5e81ab7fa1514bf3c8f655eaf7854311"}
{"id": "train_039021", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "0dacd5b63e0cb9b93f947a84c9045d5d17489128"}
{"id": "train_039022", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "90bd26758ee1b66e43eaedb5159161a36bfcb635"}
{"id": "train_039023", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "1df9915de17eb2a0b82b1a31178dcb0443cb55cd"}
{"id": "train_039024", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "9d5729b902318cf70d3bf12d4e4a1b2540ea5253"}
{"id": "train_039025", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6e26ec81b43d8e382dd96129a61b2c56e432a93b"}
{"id": "train_039026", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "da6371b5c5440d96a5a2c88d454f406685902d94"}
{"id": "train_039027", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "4c6dfe009a02d679faee03beba5662cca58f2e09"}
{"id": "train_039028", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "fd3ac7f3e2435937b6cd64ef1a18fa86f43cd48c"}
{"id": "train_039029", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["data", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "e0100b1a6c246288a06a170f332ba469d1e985ca"}
{"id": "train_039030", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["coding", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "92423650be745f8332f84de814977fcc0c8f501b"}
{"id": "train_039031", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "d49213776f545b45bb8e67aceaaf0d88a11fcd70"}
{"id": "train_039032", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "fe3dd7119fc96afab48e357475d48cd3c7931d07"}
{"id": "train_039033", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "57f0a3721e3e58d64559d4e481e2cc9f81b75abe"}
{"id": "train_039034", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "748cc9644ef5615396c4b9800198ef75f7cf2c06"}
{"id": "train_039035", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e4b5f4ba831f22caa16135baa79354ed14693276"}
{"id": "train_039036", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "a8a1db1dabfe472fc9e59a223768d258e7487b09"}
{"id": "train_039037", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "19e10bfed57bf063b6caadc9c4391af3b622f6b3"}
{"id": "train_039038", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "3311b285a31eb0eb8aab0f29cf64ede756ece082"}
{"id": "train_039039", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "43806347a4703b5030c2b966b65363043578fb6f"}
{"id": "train_039040", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "370c5d89fb1c8540f552c64daeea0d35fe68e0ea"}
{"id": "train_039041", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "95909314a78b465ef88bac5698ef2c5ce4285e75"}
{"id": "train_039042", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "a599e6dc0f173b20d0c9ab14aee723faa3d33e5f"}
{"id": "train_039043", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c7fe978ab0487b355ba94ee092d071f58ad5462c"}
{"id": "train_039044", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "25ca219fd813a5658d5c61d2afea3f0a092e4c22"}
{"id": "train_039045", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c93b2b3db404018f8fc5fe2fd33173d7ab608933"}
{"id": "train_039046", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "b66170c152182680ed647e703e8df7ecc2c9e26d"}
{"id": "train_039047", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c5cba086148a4fb2339a34237ae0e091d7cb388b"}
{"id": "train_039048", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "11792f1410a7ebf82d0c47db7b58b7460cff7816"}
{"id": "train_039049", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6f620d6a7c2307d6823780e83df700f49ed8d5a5"}
{"id": "train_039050", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "31ef3531359c0e2e11211cab360e8f5d0fe7e304"}
{"id": "train_039051", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "ae667a2589b7f711222a5efa69d24e667f6c7418"}
{"id": "train_039052", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "26bc4bf44da031211d1b19d9996400186f34f1e4"}
{"id": "train_039053", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "68f267ab99b3eeb486ebc974e1431c3c222a99e8"}
{"id": "train_039054", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d4e7f1865afa57f4f68c7803a6ac3bf0d1134e5a"}
{"id": "train_039055", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e785e16998e7f344b33400c0d953940b68d68086"}
{"id": "train_039056", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6bbf04771585ee1c630f615724e1d09a29f3a75e"}
{"id": "train_039057", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6003489d3453b4fe520a3a49a34388791013f071"}
{"id": "train_039058", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "cd4ea64a8902caf584a7d2afd4d134a24dbd5ae8"}
{"id": "train_039059", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "4adfb0c1ceca580e53e4f403b91577854d87c7fe"}
{"id": "train_039060", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "688b6529241d3568c97e57e452a1a4e2111491e7"}
{"id": "train_039061", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "6f06c5e28e27978c520f63b9c1fc2e6c57b7a353"}
{"id": "train_039062", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "9788c26098279184d8e586dc28024de2d50ffe52"}
{"id": "train_039063", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6754ec6938c389a359ae053078b69bd64a9aa791"}
{"id": "train_039064", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "352ea86bc3151223c9c15d258f12272c2b330ca5"}
{"id": "train_039065", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "684c7f406d491ff79bd777d4d715882db2f66a48"}
{"id": "train_039066", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "96d6607b5baaffde5d213e24512bb3747cd6d285"}
{"id": "train_039067", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "5160a31db57211ea02af3e7c1f0249a1b2c3db1c"}
{"id": "train_039068", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "5ee2f63f476091c3698d4882d180f25cbe9f6b15"}
{"id": "train_039069", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "540f6ad46dfc2672997749cbef38b7ef2b01d3a9"}
{"id": "train_039070", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a71e86bd47de2cb59299052465abeeade7907052"}
{"id": "train_039071", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["research", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ecc4802494ad4bfc7ec73e2e496936819a63ab55"}
{"id": "train_039072", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "19b2a5b54e87b966b7024fbc1c5ac25c21d234c0"}
{"id": "train_039073", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "9d543fdc94add9b153240a7d87e34e59247dd992"}
{"id": "train_039074", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "50d48f81befd0f12e005aee1eb93af178d79efba"}
{"id": "train_039075", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b320f1a9bbbb0607c1561cdc3dbd5283ceae93f9"}
{"id": "train_039076", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "e9f7a1f5267385da4c4a8daa3aa6678ddd5ecca3"}
{"id": "train_039077", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "e666b2d69e8047fca064f1721f3123cce93931af"}
{"id": "train_039078", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "79046f5e81eaca49b9b4ec0f837d891849bda8df"}
{"id": "train_039079", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "567afd009f5a0f47c534c7e00524a081dde27795"}
{"id": "train_039080", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "e893d91bdc2b43595991c84e86bb4611d7724fd4"}
{"id": "train_039081", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "634eb9a8ec6054ba5fb753b68c5b0383f0a67505"}
{"id": "train_039082", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "23a78313e39782e690daf518cb9d5fc2897b06b0"}
{"id": "train_039083", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "7df38f78c8bf6e90ae85375554fcc359659f18ca"}
{"id": "train_039084", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4c5090e4e543aa0759226e269dde615158c13e87"}
{"id": "train_039085", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "0024d2e28587bdd9aab797e9543b8164ccb3ebde"}
{"id": "train_039086", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "e9c15abc9eefe55c266945f616f50d76961ec359"}
{"id": "train_039087", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "f6e4a72098a6114941ab4e786947db94453edd69"}
{"id": "train_039088", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "553f5e7864fefb22e0fba08f75dba8157ad43310"}
{"id": "train_039089", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b443a45592a6ce53db27007427a36d665cd2f526"}
{"id": "train_039090", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "9097529294c5b4ef225f2881c0ec55cf4e695d75"}
{"id": "train_039091", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4e561878b179666fea17bbf89aa4704abad9d114"}
{"id": "train_039092", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "412074daf688119b150ea60e9da3c4bda8eebd82"}
{"id": "train_039093", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2d6bb58bb09421d52a8bfbde4911f071475c76f0"}
{"id": "train_039094", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "1aa9a4e30192ad62492d18e93be74c2bc56f51c7"}
{"id": "train_039095", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b955aec045354f01b9d0f8394dd2f4a8263d47f4"}
{"id": "train_039096", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "fe44ed408fd7913c3895fec1c00be4b0f1100fe0"}
{"id": "train_039097", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "81ab9745f764b1e3979fca5fa63a178be84d1aff"}
{"id": "train_039098", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "34fcc04a002e2ccd1e248b116ce06a2da7c34338"}
{"id": "train_039099", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "fe72144bcf5b844bb222fc31cf02744591a91358"}
{"id": "train_039100", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "c32e652125af2fa2f594c6552f71b1846f8a7797"}
{"id": "train_039101", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "265ad80743535cdcd3d5dd91a44b5f7e2aeb3af5"}
{"id": "train_039102", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "7d2243f90cf0d7951c74cc710bba51d3a4764ef9"}
{"id": "train_039103", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "51d67b51da9d6b64c0b73f6d33c7dbde144edd7c"}
{"id": "train_039104", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "95633fe12a6e9d42ff466659fb2b2ce1826cded8"}
{"id": "train_039105", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8cb14f7926c86dc169bc78ee25453fbbf9815ec4"}
{"id": "train_039106", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "22333c25a1ee1657dea6fb81083e093900b4ed7b"}
{"id": "train_039107", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "c973c1b4bbe813839b702164c5638a1cb9c065aa"}
{"id": "train_039108", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f44bba0cb3b8e735794ae260e84f9d73db20eba6"}
{"id": "train_039109", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "5b4a5e4aa8e8126369ad3cbdf88f40f9c8accb4a"}
{"id": "train_039110", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "32edb3bf6d0b18393a4df0ecd946b07d75bd7d17"}
{"id": "train_039111", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "3839c779f43ec5527a05bffa7113bfb2105f82d0"}
{"id": "train_039112", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5ae79b0559be0fb6dd0ce70d2031ecd54355b479"}
{"id": "train_039113", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3dd1ed2f298332daab5460758a0b2df6856ece18"}
{"id": "train_039114", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "695215fe046038d7b270e144b00976a029b385d6"}
{"id": "train_039115", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "d9b2e45fd8ecef011ff76c63bf728b2ad0b59e6f"}
{"id": "train_039116", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "e3d723e8707a381edd462b7200cf2930aa8c86ff"}
{"id": "train_039117", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c686e9b148bf580f22850f3540b84fe71b25fe95"}
{"id": "train_039118", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "95436c3a24f8b2e82249525082bb88297cd67ee4"}
{"id": "train_039119", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "eb57516477ce8a75298a364ae0954bbfa226d984"}
{"id": "train_039120", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "433c152c8fab0187ea4d5bf7e8c9345c38f78bd3"}
{"id": "train_039121", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4d182bebeeb883bc8852c9a4d11a8d1bc974252b"}
{"id": "train_039122", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["agent", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "3d12dc83c50c2e3363e806b7f6abca3598e316e9"}
{"id": "train_039123", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "01b77535e21b720d26e32a1d7e2d7f42dd6a376c"}
{"id": "train_039124", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "eca7f9f546402669909688bace8deb3eb599176c"}
{"id": "train_039125", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "021545f8506ce7a3cadbaf657214c5aeefbd89e1"}
{"id": "train_039126", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "4e6fa328f1a417265c8ea73a5740e1af22e6c897"}
{"id": "train_039127", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "2bcbd4d69d306720e17139f82efac1997dfbf8e7"}
{"id": "train_039128", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6a850cd562ff86d2a841f9738e5e4412e67e3d16"}
{"id": "train_039129", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "9e2c12a215dffccac52d1b88eccbacb5db1be5c0"}
{"id": "train_039130", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "15edfe48cee6ffe781744c65a04a29009f4b839f"}
{"id": "train_039131", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "df52cfb54d55529430fb514ba61d5c815271b746"}
{"id": "train_039132", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a6abfb8f4225a3ac57f32bbda1a802280588ad90"}
{"id": "train_039133", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0a800d6e197ce89ec4ad352588b9761fbf3fe724"}
{"id": "train_039134", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "8a722794f8061e0011b2b95cea02eb58f308bc74"}
{"id": "train_039135", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "cbeb7691dd0726feb4bf549ebde21a1b395dfb03"}
{"id": "train_039136", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f067571dce211d135a511f1298e6e924472acc8f"}
{"id": "train_039137", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "8269f02a66cddbe595d1f348a86ba1d171a97460"}
{"id": "train_039138", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "bdbc0f5f3a9710a88efbac77a87a3f8af4bb133f"}
{"id": "train_039139", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "db29ada87197ad1cc5d59ed8cd261bb3e6ff81e1"}
{"id": "train_039140", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "b198047b54f0cd663c82625662119091be871a2e"}
{"id": "train_039141", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "58ed28ad7065f821c4a0999b3522ce284d9b7535"}
{"id": "train_039142", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "0eb6c6f944151d638abc73313ed159930ef1510d"}
{"id": "train_039143", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "2a1c7867995931ef418d35e364b4c3e763785a2e"}
{"id": "train_039144", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "6ee8386353f54e2b6342e5e5b6b87a49e259564d"}
{"id": "train_039145", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["performance", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "8dbc74bf8e34b3aebae0a93a5c3dddbcbc7f9ac0"}
{"id": "train_039146", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "a5de435469925a2a2a5d0553c8eb84b6cf7168db"}
{"id": "train_039147", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "bd931fbb9e8c1b8d28ca38ff5119e719efa69a60"}
{"id": "train_039148", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "468f34fea52194dfa73c8255f557b8bec22135e0"}
{"id": "train_039149", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d7c4a92adedb36e3c3e8b1900482a99ea95d7385"}
{"id": "train_039150", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "4ff26eb15f46e68e2430964730eeb73c77a1f291"}
{"id": "train_039151", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "267af046b054dfebc6feee2009b04b50b0c29862"}
{"id": "train_039152", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a40f712e5f446c5ab15a0630bf21b86bd23715f4"}
{"id": "train_039153", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d8cdcffbd5c1d3183abbf50e4a28b6057abfccc8"}
{"id": "train_039154", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "4dcba614b274ab1cb1341ba3dbfa8dfc5a13e4fd"}
{"id": "train_039155", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "1cde996d8a4ac5f366b6978db318cecc440e0a2b"}
{"id": "train_039156", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "TypeScript", "moe_experts": ["research", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "76ada0ff613cc1b8e3222a81b118789ec4035a91"}
{"id": "train_039157", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7fac73e6c3cf048fe122c26a5e9eb91d99c83348"}
{"id": "train_039158", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "00f807037862536f5c271a3c1a619d67ef1d36eb"}
{"id": "train_039159", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "7791beefe59624490c133dd038a7871c6996b81f"}
{"id": "train_039160", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "46e7816ef1529bc0a5582557fa4b93120d2706b6"}
{"id": "train_039161", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "70fa9449ec611a023af6cc09d01ea55cd7b56de5"}
{"id": "train_039162", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "16bc5199459331040cc47180af096e7d3ea4dc0e"}
{"id": "train_039163", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "776f47b535f6ac763aae7e4f037fa46f60b4d09d"}
{"id": "train_039164", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "92cad0b53247ddd17fb2ce4ec4b23aac5d8bdd6f"}
{"id": "train_039165", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f6a5ea7f469b5a6f6750a3cc348c24f2bef464b3"}
{"id": "train_039166", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "729b801c3dcfaa7022040939a1a2db1fe1388c3f"}
{"id": "train_039167", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "bf3c1e309ab72d0c9bc16cd7d78222ecb8b9abb5"}
{"id": "train_039168", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "702a127cdc663f38aa43ef5a5d3a7b8f7d19ee3b"}
{"id": "train_039169", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "1fb9b3e6c33aff7e2811bd074fd7e1ce5f415625"}
{"id": "train_039170", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "2a5e6a2bab4b71f95e08c713647c506ef5c6d2d7"}
{"id": "train_039171", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "a7a1aca85fa7286e1a008888aac3141b2df059d4"}
{"id": "train_039172", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "062f2fb21a005c76124af63f841ac54f44285a49"}
{"id": "train_039173", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "85db00c63fffa6f80f72fb1fd5652a5752dcd4c9"}
{"id": "train_039174", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "346b82f194bd17ccc37d899f0903e1cd283362f0"}
{"id": "train_039175", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6abd4c26981ff504675a568820fa73b51019c707"}
{"id": "train_039176", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "87691735f9a991eef8493e107c2384f9a7dfd999"}
{"id": "train_039177", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "40b4e2ffdba30a8fdb0b89b207271236aacf1cd6"}
{"id": "train_039178", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e713336670c918157071488fa943cf466db375f5"}
{"id": "train_039179", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a92a1ce538a3b1090939668a545fb2f8ce8e85e1"}
{"id": "train_039180", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "1e9689e3436da2fd8dd6df896790b153e1544c89"}
{"id": "train_039181", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6d9a03d6a97049380f7ddcdc55a404a6e7788151"}
{"id": "train_039182", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "78b1139babbf69717c7204b81e35ca342f2fbbc4"}
{"id": "train_039183", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "b1be0c5f917f6042cd8551aa310238634f3f665d"}
{"id": "train_039184", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "79fd442a23a9b008548a8389439657cbf0c80550"}
{"id": "train_039185", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "147421f8723ce46fe230158a9501fe8bcf960db4"}
{"id": "train_039186", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["performance", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "9bf9a76db153f5ee82236d7ab25775d798fb1fbf"}
{"id": "train_039187", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d26f854fb8dc6d3a7e3950e4fbfe764f379c5806"}
{"id": "train_039188", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "dcfdece8466f01bd6a94d90c25c889a98f5cff1b"}
{"id": "train_039189", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "3f659bd2b40eb7884b1052015bee9cdf741aa19f"}
{"id": "train_039190", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "07278b24fac8e7f18a9a18d602dace725bb06906"}
{"id": "train_039191", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "e1bd54ff57fb316b9c12fe67cb41189a9f2dfce1"}
{"id": "train_039192", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "07be1ba0aa8985f84e97e4e6270a06c0623f0364"}
{"id": "train_039193", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "adff9804c48488839156bde9a4d7104e1b0ca443"}
{"id": "train_039194", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "86be5bae17af6e78a7b6214941269bdd2782c4a4"}
{"id": "train_039195", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "0369319836a6ce66d48c21129f7f4012fd78464a"}
{"id": "train_039196", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "cdd2b25aeefdd6f5099bb1645e2c76691b7fcb63"}
{"id": "train_039197", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "aae124daaccdab43afe9e185620d17199fe77a5f"}
{"id": "train_039198", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "d996264d983392bfab72a2ab213724f292185c83"}
{"id": "train_039199", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "a8b790ef38229872a5c916987f1e116d0159165c"}
{"id": "train_039200", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ac2b819dc2a4e9efa5196333d0ff61d435bc3961"}
{"id": "train_039201", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "fcfb965520671cb1cd4ef622af6f6605f2573710"}
{"id": "train_039202", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "26f68f4055fc0c19300e16e6e233d29d25525a60"}
{"id": "train_039203", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "ba80fa3b94fbb41260dbed8b390dde4d1c9e7be8"}
{"id": "train_039204", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "549e3dce71f03e56d9646b0fca0fed7ba7846f20"}
{"id": "train_039205", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6f6ace513882d6329358dfec53b029db796ec61b"}
{"id": "train_039206", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "bd1bdc736389d468077cda22988808b0894dabed"}
{"id": "train_039207", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2e852726f1e997a8fd94fa63d9c6df61ee4d2380"}
{"id": "train_039208", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "09c42275a440fc8fbfc93645f8c8c78c2c3c1f1d"}
{"id": "train_039209", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "0936adefc23ea299b1bc5732af978aaf223312bf"}
{"id": "train_039210", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["governance", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "b6b250ad6a26846d3e3ac50e885fc0b786600855"}
{"id": "train_039211", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "bed1ffa2ad04866c8e1780982f2875d01e7fedb0"}
{"id": "train_039212", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ebd0a44b1fdbf5fbba07d9ad0957049228d4e942"}
{"id": "train_039213", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "af5ed1cc5b98123f89087db04c7c92861a15f843"}
{"id": "train_039214", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c8db01edbf25496ad99a842357e7e56c46ebc5d3"}
{"id": "train_039215", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "b20522ccbcb561ec8a43639ca41963fe8e8b2508"}
{"id": "train_039216", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "01ab69ba6b084f4a5165c969192a6f8087abbf36"}
{"id": "train_039217", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "2fe72cc6e108036bdba9ec55cde891648b7afb83"}
{"id": "train_039218", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "453668e647333931b5fa77333920b61ad195c802"}
{"id": "train_039219", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "0587b85c760fdf59e2a3c1cd1fb9eb61a18ed0b3"}
{"id": "train_039220", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "b81f472a2b63270332ab257c069ca24733cecc40"}
{"id": "train_039221", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e2655c4077c07539fdc9ba1642f17b5f1357ca1b"}
{"id": "train_039222", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "399067b0189f378ef805b0364255a53c18e40b0d"}
{"id": "train_039223", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "d3e2b46255c27a92e0443ec63df864a84001b97d"}
{"id": "train_039224", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "1f2273108d9327ad9e8778499255bb50cdb316f9"}
{"id": "train_039225", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "ecadab1cf90b696fb283bcd65c8532e6f509dbbc"}
{"id": "train_039226", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "80afe8296b8c183597133c4e2cc45100e18b5143"}
{"id": "train_039227", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "ecb227d4ee4c85c7e1308ce6aa2693ad097ed129"}
{"id": "train_039228", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0899d484a49a1d52fcd139ff2050981c0d475c62"}
{"id": "train_039229", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7acc8fa431e884ae3866d0734198dcf21eb22f63"}
{"id": "train_039230", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "4f3d8c33719960b31c494f268aff793db1d43512"}
{"id": "train_039231", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "fcb993d46e724c83d5df613ca81c912225043ab8"}
{"id": "train_039232", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "32475ce6d535cb77cbcf5ff01a2666d119e5c3f2"}
{"id": "train_039233", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "0146d43e27a02c79f6669d881a73462fe84072f3"}
{"id": "train_039234", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "bde4ce34ab768bc0cbd21471a3206e069a5103fd"}
{"id": "train_039235", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "a4314b55238003dd4f8784caab5e8d76bba03e86"}
{"id": "train_039236", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "a6aeb51d020757ceeabfff0af9d3f70bf442d3c8"}
{"id": "train_039237", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "dc46fc710b9b6726d1dfef779baf0e37c2b871db"}
{"id": "train_039238", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["security", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f17fa60bf2215e0ed155132abffbb81cc7f41c7f"}
{"id": "train_039239", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a3b5a429817d783bbd8e60c65f4adc450eae838d"}
{"id": "train_039240", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "a4e8aef6754fa07697ee9735c3e3948d9cefd40a"}
{"id": "train_039241", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "da2b0d59545bfb558d47da91212753fb6589a17e"}
{"id": "train_039242", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "f05db064c40489095627d34be77ba61c8eed31c5"}
{"id": "train_039243", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "787b2b00e935d908bcca36a4fa3179bc9980f592"}
{"id": "train_039244", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "281ce2e057fa0feb172f3e90adcade717a295466"}
{"id": "train_039245", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "22980b0fdfd5efe75a15f2a1f88c0cf7ed0eaf94"}
{"id": "train_039246", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c4e8b8482179e775381d9108c21be514e63f4e1a"}
{"id": "train_039247", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a2ff449d7eec8ab52ac50359b3d54f0499cfac6e"}
{"id": "train_039248", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "9c896f532384e5f56aaa5f032fc89f6a9b27cbcb"}
{"id": "train_039249", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "a759f45423117992b376a14ea93d611c5f298d3f"}
{"id": "train_039250", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "9f2b7edd9f3088877f964e48433d0f54ff4eedf9"}
{"id": "train_039251", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0f0cb0d4ee14968dc53474d9b73ff359a955ca73"}
{"id": "train_039252", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "2b0a299c5cb1f1b7f35635979aafffc4622aef3c"}
{"id": "train_039253", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "ee70b35cb21ed6480cf136821cdd9a875575801b"}
{"id": "train_039254", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "505762040219f1659b725a342e8ce863879da074"}
{"id": "train_039255", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "0ac85336095dbd141f9afddf3f16054784799e35"}
{"id": "train_039256", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6b9f5f817311b00d72215e2c6dcbc81d57cd8d2e"}
{"id": "train_039257", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "a4dc36361b5cb6b1f04f7e42e6583cb43dc2afaf"}
{"id": "train_039258", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "22366c088539c6b55577fa988e4ef910194fdf20"}
{"id": "train_039259", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "70f29c0081d7779537121c9ac68df109590fe14d"}
{"id": "train_039260", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "b3dfcc143d09174b0d0b06ef6d2ebd72a5b6d533"}
{"id": "train_039261", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7b5043b67e49f8d3031b8ad12ef730e9f4c0a142"}
{"id": "train_039262", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "27dfd2ab45d5f0ad233f564a675853eb33f080fc"}
{"id": "train_039263", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "08b2f3af640ecff2aa32b6b50aa00b307ddd7ba2"}
{"id": "train_039264", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["performance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "1d7f91ecfe58239780ffacd6b86a51702a304508"}
{"id": "train_039265", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "800dd7c3b3db9013ba9c1be5c01aa34027e0d2f0"}
{"id": "train_039266", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "32b5af05ec719064a35b3d03747508d0d93f465f"}
{"id": "train_039267", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a2ce7c08882776fd71aeeab96449e271b4ea9139"}
{"id": "train_039268", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3501f9ce735bf72aeb5904beae896a5f6fdd9abe"}
{"id": "train_039269", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "61d5e50e320a5131a20292355cecbf1d1524f3ed"}
{"id": "train_039270", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f70291d298f608befaabb2ad3751b536aa116dfb"}
{"id": "train_039271", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "54e78e90ad14f9a9f570670910416d8ab5619a79"}
{"id": "train_039272", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "47aed3deaa6ad54fc3302e0fa5c4693cbf131ce1"}
{"id": "train_039273", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "be92119acecfa3ece4f6750e0c64ba5c80c35da3"}
{"id": "train_039274", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "49eb34e5547abc98ddbb1d58cb7bffc71819f64e"}
{"id": "train_039275", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b66d9c01ea694b985d632ca5ca7609231e09f46c"}
{"id": "train_039276", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "cf3743f546d634469b21777013e8a292d2f96eb4"}
{"id": "train_039277", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "b07c397dac5f77a711e915c58430dc7d02cc42f4"}
{"id": "train_039278", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "373805431061bb71ca93b7faff4d7366130bc299"}
{"id": "train_039279", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "86a07249eec76aa8519242ffd7df03c55d34e27b"}
{"id": "train_039280", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "2d261cf0bbd9f20ea7cdd77435e7d4c0893baa66"}
{"id": "train_039281", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "63a546822f0353a64856b85e0bef244a44e922d9"}
{"id": "train_039282", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "5bbc5bd10640d01d75cb386f49c51902788ab640"}
{"id": "train_039283", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c9a9dd4f9ad7d54cfb917dfcb21bc9526e43b335"}
{"id": "train_039284", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "7668183fef35ed0c2838d86c8c38b59225100651"}
{"id": "train_039285", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "e9d29b7bc90f2d4b722d05295e1f6ad55e459b02"}
{"id": "train_039286", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "4369397aef07c4cad1c6eb49058d373f0ecbd391"}
{"id": "train_039287", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "46eb6a76cf554c05882296a4bbbad087bae63d2a"}
{"id": "train_039288", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "92795d648f46a8a00c20e6ff30f580ea51624fab"}
{"id": "train_039289", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4686de4778771f1d7cfd229a1bc6727905a982fd"}
{"id": "train_039290", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "136bb8fcc51a9a6e3f38ef0f7950fc31e518de55"}
{"id": "train_039291", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "209e977bd40b3392019e183d78948db3f8c07eba"}
{"id": "train_039292", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "21392d7ff6b23bdbf01d596aac75168ad970a7cf"}
{"id": "train_039293", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "78a6bdfcd1667a64fc7bf488f4ca05f07793171b"}
{"id": "train_039294", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "501a795eb274bcd31122ad585fb3c39bc4ec20a5"}
{"id": "train_039295", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "2dd44c2fa6dccceb79bb24fa86cf5ab2d7027029"}
{"id": "train_039296", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4bec171fa6003e0745ad5891c415ee781a63e8a7"}
{"id": "train_039297", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "8f6342c3e921dad31cc03eff7f5e94e3211e4c90"}
{"id": "train_039298", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "114e2828f742477bce74698408d532a482dcac9b"}
{"id": "train_039299", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "357a7ee954cc888d18082a1da8a80cc852fbaab7"}
{"id": "train_039300", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "24c2d71bba1b0e66d1c238219b75e4f17313d542"}
{"id": "train_039301", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "626cb477fac7e47a44e52258cee5b9d808e78e62"}
{"id": "train_039302", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "26e65aff95d5c10c38c1f658a460295577b68994"}
{"id": "train_039303", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "44fc19aa0e78c20b993545bc6829df0037563600"}
{"id": "train_039304", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "41ed841b6dd7380be6355ce8278b3eb5af1ffbe0"}
{"id": "train_039305", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "eabe69670a2c1c55c04cc9837f93a85befa013e7"}
{"id": "train_039306", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c1cbe0c2343dfe29e159517c96b4070fd106b693"}
{"id": "train_039307", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "dddd9657f373cb385c0b43c04fb3d1ec50605cf2"}
{"id": "train_039308", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a9aaa3ebcfaf066d5eb6c0a78c72a2399a2cb8c8"}
{"id": "train_039309", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "3d1bbb2e53f47b4ed6d9bc9b58c4a9ef8c1dfa59"}
{"id": "train_039310", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "797f524e04a12fa54749ab4117b6ec5f38b19ca4"}
{"id": "train_039311", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "c0f9fb880b04f71a41bc1601690ed479cd25fa7e"}
{"id": "train_039312", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "eda6f45d1b9b166c67442e8c55e2e389b445609d"}
{"id": "train_039313", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "36084513ba5e694406a21ce4eb43508c11de9f5f"}
{"id": "train_039314", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "ac5709f8113c533fad4e6c749acdc2d5e9bc5ab6"}
{"id": "train_039315", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "35db3d223e108e90fd6f8380e6a2a19319e96c4a"}
{"id": "train_039316", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "77c883fa3673c073f1303ddb5dde3ed94fcfd311"}
{"id": "train_039317", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "88787acb6b5dc2d5fdc30ef7ee76285ce21debd6"}
{"id": "train_039318", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f8a1b36018ffc57362df6985513a8c5e24b6b193"}
{"id": "train_039319", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Go", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ed3c0e0e5219424ae82d92756d317fe035e3fcc3"}
{"id": "train_039320", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "43867e2b470d5683e74ecfd0e66456d338c1ae16"}
{"id": "train_039321", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "681c6b7654018a4951d6e07e5128db0c41a95d83"}
{"id": "train_039322", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a9368148494ba299135a9f2a9225726377ee0d4c"}
{"id": "train_039323", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "892ab30006849b5c8fc8beb23b4da015d2dab887"}
{"id": "train_039324", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7c9730a91441d7b5eb4fb4b76dc992988a6fa276"}
{"id": "train_039325", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "9a39eb530c2fb14ded88901254b180d92046bd92"}
{"id": "train_039326", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "d9f3cea41eba3e92a9cc3f406391828e8ee57051"}
{"id": "train_039327", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "8a6bdef1422a98cac26e4a84bb5e3f9a232e202b"}
{"id": "train_039328", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "cb263caa8fd304e12c4bd40cdc244e9680221046"}
{"id": "train_039329", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "25a3d263fcd94ffba80d7a42a82a7a9e5b02b827"}
{"id": "train_039330", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "36c0378dd0a3eaa87c7b76f0242d4d634618822b"}
{"id": "train_039331", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "680ba1b60a2bc8a764056273849b0e7f94c40db2"}
{"id": "train_039332", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "91a640ab1c28ab9df299acbec886a5ef3572d34f"}
{"id": "train_039333", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "791063fd04577acb96513ec2a7b7ca53ea48c4cb"}
{"id": "train_039334", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "e63432baee998c63eea02f9863342dacb9837351"}
{"id": "train_039335", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "95be4481d88bcd0eaa80cf9b30b60eb2354be995"}
{"id": "train_039336", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "6c3f18a6060e3b4f98f425bab98ccecc1679c491"}
{"id": "train_039337", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "cb725a3f9203a7ec9766e512fe6a239da55e5100"}
{"id": "train_039338", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "81a48c1bf2d9c068f47104dd98f58f9f0730e1c5"}
{"id": "train_039339", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "7de35f94fd00ddd5d47264b59b237a462cf5eeaf"}
{"id": "train_039340", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6746fc9eb5c375ec2e3b5e9c1a4e35c6c570a7da"}
{"id": "train_039341", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "4789d3607ef236c98beca9d29e097c3c96efc596"}
{"id": "train_039342", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "445846aa567d4f2c66f4fbab44a31ab114e946ab"}
{"id": "train_039343", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "13d3812b1f6b4ba01379b2457c6a2eea40b53a87"}
{"id": "train_039344", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "7ce100daf917d1c2288a4196e44754518ac6569a"}
{"id": "train_039345", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "8ea0355905e71dcb2520ddf84f7f076ae4f6329a"}
{"id": "train_039346", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "98bb04ded432b084b1ea0eabdfbc582a18124daf"}
{"id": "train_039347", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "12a2cae999e41ad0f183afaef8279bfb98295126"}
{"id": "train_039348", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "f8cb345f2390093606cb4b84c38d212487052250"}
{"id": "train_039349", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d713b201c0dcf6a6585e02d47d72db22ed805b8d"}
{"id": "train_039350", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "b7ef183b50381f5085d45d921762e8a695b4f1f3"}
{"id": "train_039351", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e44878f6c76ce75a38b91b6f9a21a9def6c63bdc"}
{"id": "train_039352", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "64eb109fdf98e1d77b8712fe18200cd88e480dad"}
{"id": "train_039353", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e26f15ed58edf3eab2403402c0dc5b115d48823a"}
{"id": "train_039354", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "bc233da0b8b6f4845e22a38275bbe41e5e217ee9"}
{"id": "train_039355", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "dd1a59433d0850c21f70ab5c03cf27e332da1dde"}
{"id": "train_039356", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "f823ff5166744adc330681c19909810581d16437"}
{"id": "train_039357", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "592c9ac371ff11ffb7f0db0a3d4b7b3c1619c4e1"}
{"id": "train_039358", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e44878f6c76ce75a38b91b6f9a21a9def6c63bdc"}
{"id": "train_039359", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "2246bd32d11285164802ae185b9d75615c1fdcda"}
{"id": "train_039360", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b2de401305e138670eb19bbc55af752b7d02c892"}
{"id": "train_039361", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "76731075d4eb4e5b69d996eb6f85f8787280b96e"}
{"id": "train_039362", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e4d74d17d3abe8031766bf607feabde9001fd929"}
{"id": "train_039363", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "2feddf3eebe3f97d2b961b6fe8064f006e347960"}
{"id": "train_039364", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "23b8cd57d7e38f1140a2395ce0da2e58ee9261de"}
{"id": "train_039365", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "9c6446545f21f0ee0c952470d6f4c8ba9c7831b9"}
{"id": "train_039366", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a1750a21a9eeab8e8b89c7bd9323e30eb38ed47d"}
{"id": "train_039367", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "a04c21fdb7a770cdeb5c23bc1d25810519647b1b"}
{"id": "train_039368", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "8e44a822d6e30b0c81df18fcb68e3bc9e9eae1ac"}
{"id": "train_039369", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d4d4b2070aa2c81c2c53bf0620d63bb209cfc5a4"}
{"id": "train_039370", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "42d6cd43db389a38686b77b3619af5e26adeb256"}
{"id": "train_039371", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "02246aeda5e2de81a1b266c71c22c595f3597841"}
{"id": "train_039372", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "5232abfb9176a2c13741df0c3efc301fafbb84fc"}
{"id": "train_039373", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "1b8a73c7523f5696eeae40f76e57341c3dd1fcc2"}
{"id": "train_039374", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "6003489d3453b4fe520a3a49a34388791013f071"}
{"id": "train_039375", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "248ac7bfb58d5c55dff1d96fc11c5182d553ba2b"}
{"id": "train_039376", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "3296aaa34798b0b9c0f0600ef84c913cb824bf5b"}
{"id": "train_039377", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "529b8c9e53d3941dd28849856141a3c5d31cd724"}
{"id": "train_039378", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "7e4e00facad69e930dbd4d4afef71b7c9a8e6fa7"}
{"id": "train_039379", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "fbccf8bd103b2e5320ff5a14849bffd9e078f507"}
{"id": "train_039380", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "983873539c0f37851c8a3dc8f773ac393b887a6c"}
{"id": "train_039381", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "e9b39cd0c7dedaecd8960273646747197d086c5d"}
{"id": "train_039382", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "af683eca50879e67c6ac8be6ae6578fda552d566"}
{"id": "train_039383", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "91b51738183f1022d4dd902317752a9179966705"}
{"id": "train_039384", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d6eb2ea2d3f9a52271ba514d2a9cc4b0c80f52e3"}
{"id": "train_039385", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6ef328a86a1f8749784f01fe41f154536141e5e3"}
{"id": "train_039386", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "2d17dfb27c39f75d5cc5e5c99034c5490f91b02e"}
{"id": "train_039387", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["agent", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "3841b94b3575664e1ce5dd04e88c267eac01e250"}
{"id": "train_039388", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "027921a55bba5b1543ab3b3b844e74340b5e9be0"}
{"id": "train_039389", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "c28c5a16c8d31395faa06809d1f91e483fa1b7ff"}
{"id": "train_039390", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "ab8b787a0870cff91563f74764164aeed28fb7e9"}
{"id": "train_039391", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f9f19329c23e8ad6ee07372aad38ffaff75c5eea"}
{"id": "train_039392", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d5e257e0a566b7ccf68ff9dc7789b8d5c6813769"}
{"id": "train_039393", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "1ac7733071ff97ba3d2e5656e02eb4103249a9a6"}
{"id": "train_039394", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "57e1a2f18f725d2cba3d59492884cfdc6ec7920f"}
{"id": "train_039395", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "3882ae2aaa6f360bfbe0b049521b09a478e33dd7"}
{"id": "train_039396", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "8735b2b6ddddc3ae2942f0fd595fddaeab147543"}
{"id": "train_039397", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "464e07d102e012419e42235503cdb7c6299f748c"}
{"id": "train_039398", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "31868f0c6bcf32651ddd3d8b97593e0506d97414"}
{"id": "train_039399", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "511ebcbb1de6223b6d005ae998eb1b216f30ea44"}
{"id": "train_039400", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "e75b9c8957cf2a544567f6130f68ab3c1bf4bcbf"}
{"id": "train_039401", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "d0f94eccf9bbe0a2814d91c40fe914eb668ff43f"}
{"id": "train_039402", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "58e884f1ff82d109a1c11c874939631dffd320ec"}
{"id": "train_039403", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "47b19b6ca4574d84debed2cb4d71db787e6a2fd8"}
{"id": "train_039404", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "402dcac0b570ad34a70aedee9992d17af6307758"}
{"id": "train_039405", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "429626b4a4ffd06edef6779e7ab2cc272981015d"}
{"id": "train_039406", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "f5d0fd588dcfd2bcab65e4b8c474e419d21d0242"}
{"id": "train_039407", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f8289300c623cd2281fb8c758407aaff2ba4f920"}
{"id": "train_039408", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e17e15f608f063060b39e438e48abafdcfa5c693"}
{"id": "train_039409", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "9ef72a8fc949bc77038074376d01f315c4e03ce1"}
{"id": "train_039410", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2d3bc534426984756cb2877672d38d255773d929"}
{"id": "train_039411", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "ac85fd65dd8b20a56cb1f8fb0c86b6007e1a86a6"}
{"id": "train_039412", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c526f9f97ad6ed6d979bcbdb977a3299c7968ae4"}
{"id": "train_039413", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d9f377d764e6c5d50041727bd61f903d816184f7"}
{"id": "train_039414", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "4983321e57466d045d21f1d285519b5d12ae47bd"}
{"id": "train_039415", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f7a9453046b3de095a11dc9f6324bfda03084f4a"}
{"id": "train_039416", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "8f981ad92523e7c38e2209045e2449dd10c06ff3"}
{"id": "train_039417", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["performance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "be198fbbd5051a98696a46e1e9879dfa096dacdc"}
{"id": "train_039418", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "78ba56a43af2f3af61424cf9cc8424cff79c329d"}
{"id": "train_039419", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "ca20ae9c072925546008ef1ab7c5b8a2a28f4508"}
{"id": "train_039420", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "657edce5907f002aa74299765fbb59eea980b98d"}
{"id": "train_039421", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "15f315a25651cf9f29bef0a9354e9bd998891775"}
{"id": "train_039422", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "b2cfadb1c247fefecf69d3c9ffdcbdbc259f4ebf"}
{"id": "train_039423", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "adfcbe2f3e962aa6de4ccb9e4f9e91f07123f8d0"}
{"id": "train_039424", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b92a89ea53a982057e7669c3e76ad1d9ee7feb94"}
{"id": "train_039425", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "91d01d777753e9f9e1d1ac132127f40d9156a365"}
{"id": "train_039426", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b0920636e701b9f04a7fe50b49051bcfe8ef3364"}
{"id": "train_039427", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "cbfe290204d2a70997084b5ebf0c3b561393192d"}
{"id": "train_039428", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "9f336e8f2c21db553bbada54030e625a2f36cb81"}
{"id": "train_039429", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "5260dca1054081a3abee9ff16a6593168c2663c9"}
{"id": "train_039430", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "1220b7b72475d6ece888282b3bed0510fbaae575"}
{"id": "train_039431", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a9f49fc9b98ad1d8ecdba049e785f628c3b8d063"}
{"id": "train_039432", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "8c16c93e640c115c3ecada714c3f155bbff03fce"}
{"id": "train_039433", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "cc51b8bc1539e4a7a9668b824cbe452b5eda378d"}
{"id": "train_039434", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "11c1d09e71b4877f31d4c13e8a4521886d95c34b"}
{"id": "train_039435", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b16b2599ac3901d62b50c44b94bc174878a60f99"}
{"id": "train_039436", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "81ab9745f764b1e3979fca5fa63a178be84d1aff"}
{"id": "train_039437", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "7eac6b4c117ba4b3d6163547ecd115ba4a7318fd"}
{"id": "train_039438", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "83a491232d63fd369bcf000931d4ab515fdad54e"}
{"id": "train_039439", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "dc3a2d58f7fb0baae7063f1f00cef9ebb4d84a40"}
{"id": "train_039440", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "8a1c7efe5cfe0e05251e2ee8b7d372046e6a1bbc"}
{"id": "train_039441", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "995bab96ed6356da50ba713d22c6ea16ee1c6b4b"}
{"id": "train_039442", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "60b40ffd923f717c986e4543fef069e83ad7a8ca"}
{"id": "train_039443", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "bfb59cef424d18b6ccee9ee36abdf8d62044c1bd"}
{"id": "train_039444", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "02b9005eff73bdf3facb37999aed2c260a018b41"}
{"id": "train_039445", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e038b87808c5417ab195b990112885b1eba5163e"}
{"id": "train_039446", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "e18bcf62f9d7e471fbace23821b7325082d3bca9"}
{"id": "train_039447", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c29e027e6748c765b044465edd99b629ccb0f9e1"}
{"id": "train_039448", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "fee5f95f6cbac1134747a93509e0e3c326707e4e"}
{"id": "train_039449", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "fb413e3e3e5984a9a39d4dff5d7039959e2be90c"}
{"id": "train_039450", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "2e5b131c399858757ce13de685815b2930d14099"}
{"id": "train_039451", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "8638447bc76744dff7b6957a7e56976f4d7fb9c0"}
{"id": "train_039452", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "87c6070c50b4cd0799d0a67585f4b49e73c8a9c5"}
{"id": "train_039453", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["performance", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "4012fb4a145746422fa4a20e1974b6999e444a45"}
{"id": "train_039454", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "0066417c3f0ff676c89e26dc65b725ff224ea9a8"}
{"id": "train_039455", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "1abd918b653e37231f793ec9c1bbfd0980816b90"}
{"id": "train_039456", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "f8456ea9470f2f3fc1fc2417abdad50a76cc3c6a"}
{"id": "train_039457", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "e03e16aa7fc0752e9dd0e9329f3de91d4b0bc7db"}
{"id": "train_039458", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2fc5a81faf30fbdd24c1436ac9e00e4a98199712"}
{"id": "train_039459", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "6c2904f9d8d548822f78ae61f9eee9b767c1021e"}
{"id": "train_039460", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "095c9de977249ecea23b0e24729cc9aa331c8746"}
{"id": "train_039461", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e619f1a65303b4900e364b5f09182dc2c1eb9ee7"}
{"id": "train_039462", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "601582afbb75fe3d919d91d00a9abaf42d24f4ce"}
{"id": "train_039463", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "c2745fa2094a76079d86f80637fa50622731fdc8"}
{"id": "train_039464", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "3bd4d365b62cfe698081b2aeb42c34226e3bf620"}
{"id": "train_039465", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "08165c0b24e62ced48f1c04a2850d522cb265e11"}
{"id": "train_039466", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "89f5310c451e4d92e059c13e06f5c9cecf30f2cd"}
{"id": "train_039467", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c6b391fe48c6a86a6405f16b1f79a1acdfcfb11f"}
{"id": "train_039468", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "07a2af5e8b18a939d7ae2b54485017afbc65257d"}
{"id": "train_039469", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "b1be0c5f917f6042cd8551aa310238634f3f665d"}
{"id": "train_039470", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "576c73596febc848fc6e79f0d224a1200c5c7553"}
{"id": "train_039471", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "097f1b175cc7aac816120becbde62f70f8909fa4"}
{"id": "train_039472", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Rust", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "f4ead8d605aac43c4b865467f93baf08a11809fd"}
{"id": "train_039473", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "53c35dc0bb19eef8f3dc8ef0ffad89c09356296b"}
{"id": "train_039474", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "030d9f74e16bf4fec0f057963f5ca770cd88fefe"}
{"id": "train_039475", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c7e6f6eb9695350371d32021c651c187dc2e58b8"}
{"id": "train_039476", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "502953ad31b3db6a48ade2181204f6c47f82ea21"}
{"id": "train_039477", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "29db0db86b7e26d6aea894f70cc5346a7534f5b9"}
{"id": "train_039478", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b289c17d19198735222cb6a0a13f4e967d2feedb"}
{"id": "train_039479", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f3ff892ff4288854f9e4c63406ca90a1aece356a"}
{"id": "train_039480", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["performance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "0f7e72a07ff727af8b145aa3d05f9adea569e5d6"}
{"id": "train_039481", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "1ea828bb4d69db9d61e8b3b2bec957b8752b32b1"}
{"id": "train_039482", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "dfe4452298469a29c2ca2fb7d0655605afbfad73"}
{"id": "train_039483", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "050f9ddfe46d7c31ab2fb14f525ad7860bfb957c"}
{"id": "train_039484", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "284629f3287b9a00f434a288a58782b81ce4d01c"}
{"id": "train_039485", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6cc7919f1175604bfebfbe845a0a15a0d292f2b4"}
{"id": "train_039486", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "0bc8f65b8f6d3a42e7c12a21ec619503c2666330"}
{"id": "train_039487", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "b0920636e701b9f04a7fe50b49051bcfe8ef3364"}
{"id": "train_039488", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "91b26a41443870e5bc894188171d4e4fe4a0331d"}
{"id": "train_039489", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ec77b5a3f064de960ecdb887f54e967162af1072"}
{"id": "train_039490", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "3b86152527e153e1bd92cad0c2e516043f8aa02b"}
{"id": "train_039491", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "aa6c26611918924a00c7d427e89fc3c53d62806e"}
{"id": "train_039492", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "8cf30d80328ced0cdb770fe66a41ab1fdfe97439"}
{"id": "train_039493", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "9143af4c489cc91732db0178ee90d0fbde5a9b2c"}
{"id": "train_039494", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "73bdc2620982e29d0d6e328a05ac81472ab79cb9"}
{"id": "train_039495", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "d26525a5f9ed4e9b79a7693c804feac823d2a935"}
{"id": "train_039496", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "4be1ea9b28831e02ba04281afc9a1b5474b92a5f"}
{"id": "train_039497", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c448433a19aa62fbb91a7a3c5e2828f5ce73aa20"}
{"id": "train_039498", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "6ecff23d29291126ad8f90df5035b29fca07e247"}
{"id": "train_039499", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "91e472a956897c06a3debb098a0b85053273bea4"}
{"id": "train_039500", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "af2d9796b08467fabd16781ab7e4e92ffc3d067b"}
{"id": "train_039501", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "162e29e2a6d4d16c19d93c3b5422a7a556542ca8"}
{"id": "train_039502", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "e3563f395dfe2ffa13ff30d6ab3d155e20ea2599"}
{"id": "train_039503", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "fd4c6afbc6c48b854697e06e8be1d75f013701c3"}
{"id": "train_039504", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b364b4cfbff735f923d7fdab7c2315141728c20a"}
{"id": "train_039505", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "cff606ee95bcb4fbaf9bdad46b8177bcfe6a3809"}
{"id": "train_039506", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "447326527ba2869e239cb9c6fd5c0e80431a7df0"}
{"id": "train_039507", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c614bb0223a2a268b5baa6cfb208de5c34ab4f35"}
{"id": "train_039508", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "48f8f5ad05ea6be84af85a466b23d972bbfef3cf"}
{"id": "train_039509", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ba56e98f52fb9359da1fe6f2d3bb1f8ad0c1a21e"}
{"id": "train_039510", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f2ec5841925e48933d7a8d3253e2ecaee79be49d"}
{"id": "train_039511", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "028bbfd8a485a0d3b905582ad7724966a0cd91e3"}
{"id": "train_039512", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "1689a552b314ac86bb72b5c7b2bfba1fc0c7ac89"}
{"id": "train_039513", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c680235903a664e96495f36a169bed14b2cf9e71"}
{"id": "train_039514", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c2641126802ae07e7cfce1ebb62670d3808f9da5"}
{"id": "train_039515", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "22a9387af765db5ee0499fb04f6e856c7d67fcbe"}
{"id": "train_039516", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e81cb2917754a948d725b58b6cfe1da31885a4dc"}
{"id": "train_039517", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "2caa2abcbec8986568abe21e085af64cd06e2911"}
{"id": "train_039518", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["agent", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "c9d6dd78e2500010a6871c2da8adbf3cdca2908c"}
{"id": "train_039519", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "5a100ac4901a0a3d76502819ef87cfc21ef52d88"}
{"id": "train_039520", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "835b8c9ad673c1943c4d903fdcc468bdf1a1ea90"}
{"id": "train_039521", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e025b72e35720ced9e61924bbd372208e5ee4b41"}
{"id": "train_039522", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "32d551228b9589dd60086f1aebf9755669cc9587"}
{"id": "train_039523", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "6c88bcbb846ba50779200231ae98f1b41d6ff450"}
{"id": "train_039524", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "07b918e94e6c360db3ebb7e36e8ece2ac406427b"}
{"id": "train_039525", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "5a4157accac8c0bcf30f8c708015aed8cd7c73be"}
{"id": "train_039526", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "56a0059adc06dc0afd3bce6e0bbf9d21381c1cc0"}
{"id": "train_039527", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "2d9c7b7ee5740331d6ef878a10a819d7685f94cc"}
{"id": "train_039528", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "67b32b23e86cf645635f4c47c2d789fcb6e2d2ae"}
{"id": "train_039529", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "e0a52c61a0b7261a746368b732fca9062585c27d"}
{"id": "train_039530", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "4160b76c601f90ab5aab3b5c4f652a8b3bb8683f"}
{"id": "train_039531", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "518ad6bbe761f532e1faa800c652ad18e08d3c9f"}
{"id": "train_039532", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "6aca16f46e183b1481b9eba1e7481ca73aec8de1"}
{"id": "train_039533", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "46465a742a79c4b57c9d9d5b6d7c2039de975a23"}
{"id": "train_039534", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f7c731e37880a7172ba486a969dccab1bd258967"}
{"id": "train_039535", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "8cade80496aa66e910361e277c78d3f3b0d90ecd"}
{"id": "train_039536", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "9c93c8e20bdbeb64fdf758e0981400268c7ca37d"}
{"id": "train_039537", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "9d91a2e1ecd2b4f9876ccc085b4f8db3dd25efd5"}
{"id": "train_039538", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "9fd55bfefd2fd8482807c69bd7a706ccea5b05dc"}
{"id": "train_039539", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "a3b5a429817d783bbd8e60c65f4adc450eae838d"}
{"id": "train_039540", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "5367999ae036e2b9dc0f3b61d96c2bba3aa91fca"}
{"id": "train_039541", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "e302b1fe904a059294e575177aec11fadc3d65ea"}
{"id": "train_039542", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "44500a241d01ed4b386f2ff8948d7f670f36cf66"}
{"id": "train_039543", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "769d0c8d75517240988e493a1c331aca5589a5d6"}
{"id": "train_039544", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "39eb8e0e09eec0761a08af8c4caf27cde59e157d"}
{"id": "train_039545", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "45d8975457e32eac9dcc9876fc6931341112ef98"}
{"id": "train_039546", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "dcd12e1af010989330b632ac295aae5aaf7a7e6e"}
{"id": "train_039547", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "74051ad0e13869014310bef77457be593e9d26b4"}
{"id": "train_039548", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "9441d2a9a1e9fd02b65386ccf065cddee7e58f57"}
{"id": "train_039549", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e7015c89be1af488babc02a1446a1e321f73de5c"}
{"id": "train_039550", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "841b1e1de15ab0d96222a22ac2cd73a47725008e"}
{"id": "train_039551", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "f2d0fdef5e8dc1b22983a56343fe59d006ee6459"}
{"id": "train_039552", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "85dc3fdc6e0b3efdb180d176082b4c10b5a80c46"}
{"id": "train_039553", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "85af49378f23a15ed73038618c181f7f9902c44f"}
{"id": "train_039554", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2ad8e5f3f0d1062d31d5a769317d90df2fba5ccb"}
{"id": "train_039555", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ba498b165be20b28b32d9ecad2517822a80853ba"}
{"id": "train_039556", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "f64662f9c2a76283ecaabedcf7e4a0932f2a46a5"}
{"id": "train_039557", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Bash", "moe_experts": ["data", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "86056c4caa7412abb7246293b985975235be18cf"}
{"id": "train_039558", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "eba099048681b4a682822641b0a3f621409eb5a5"}
{"id": "train_039559", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "0d863454aaa613ddd123d5fc683d3104b6e51615"}
{"id": "train_039560", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "2af0de016814ae23627e77e07362a9fe0ed64fe1"}
{"id": "train_039561", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "3e78e9e1a978c438aa22f7b89206c9090890932c"}
{"id": "train_039562", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "31868f0c6bcf32651ddd3d8b97593e0506d97414"}
{"id": "train_039563", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "1015d0e6d908184476287c12baa03d07515e4a13"}
{"id": "train_039564", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "501a625b3df8b9ba34b7120a815a31743ff67cbd"}
{"id": "train_039565", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d54f25a8b56feef01be01ecfc8d5417fe125433c"}
{"id": "train_039566", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4cf5a6d31512aa1bb85cb4d473975bdf7f9bc4d6"}
{"id": "train_039567", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "987e2e0a1caf514ad0972523a60dbace71a3aa2e"}
{"id": "train_039568", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "e84aa149adc67aa6a9f9628161aea034544738a3"}
{"id": "train_039569", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f018518258480b0c96e94387d5850785af2ab9f4"}
{"id": "train_039570", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "0f57a99da9464b68184c3d118a5ae7a0e4dd2041"}
{"id": "train_039571", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6d2969710539fd62637d9336a1608427eb96503e"}
{"id": "train_039572", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "1ba1cb6a0a0ab341833c2f503716c42ae9e85c35"}
{"id": "train_039573", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "f9d2a9103ac9b8fa29ece6659fe001f3728c8748"}
{"id": "train_039574", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["performance", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "7f29cac5703b8e51f56f140e458a7da405208472"}
{"id": "train_039575", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b25439f94471459472a48851975e7b0a9887362e"}
{"id": "train_039576", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "e50b372d2b3d03ce67e0de51da50f25a1e514041"}
{"id": "train_039577", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "897f9657b5e5f7af94d197755ba53a68c8bec7b2"}
{"id": "train_039578", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "e6e9c9179945e1628cfddd493e4ffc6692e95b52"}
{"id": "train_039579", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["agent", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "fe668e7039aeb117a35a61995fe9a758ac98b534"}
{"id": "train_039580", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "5de89075e67b63d2b488bc966526fb898aee2e7b"}
{"id": "train_039581", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "2784fbd341d635b6d1d0f27f4da8b46577eba709"}
{"id": "train_039582", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "49f999c59e139a647aae2e9196788b9b0660f40c"}
{"id": "train_039583", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "fe3755f9e27539b853bf9f161d75d297757ef964"}
{"id": "train_039584", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "ce566587b46d9dcdff60738c1ae45eb3cb4e0200"}
{"id": "train_039585", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "5fdbd84f5662cf4f43722caf9cc782cec7eb3c35"}
{"id": "train_039586", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "7df8bf4a50b3fa994cc944e3cff29a26cdb18ebe"}
{"id": "train_039587", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "e4bad010f2c05a59e70c187728d530dc4ac5b6a9"}
{"id": "train_039588", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5232abfb9176a2c13741df0c3efc301fafbb84fc"}
{"id": "train_039589", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "939c166a1e4eff1bda26f894bd68a5fbba23dc5d"}
{"id": "train_039590", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "359190a92f8786b30ddf95f84ed01aa439a9054d"}
{"id": "train_039591", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "7203ffb48733b2e3b37c6666449e4ac8bd2e0ae8"}
{"id": "train_039592", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "566445605b78bc782ba57ceafc6b696bf6af2f4d"}
{"id": "train_039593", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "07083013a4241404ef6327746d239a387627a967"}
{"id": "train_039594", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "c92b42019431d00f7a8ed019ae9672c217e48c64"}
{"id": "train_039595", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b8f6aff5b7915331e348683d5d2afb4593715937"}
{"id": "train_039596", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "84903fdf9642cff3bff2017230dcb5e7409714c4"}
{"id": "train_039597", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "3d39381c7e38efdcd3fff8be7440da05830e10bb"}
{"id": "train_039598", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "559a01edd34635576b5305368d0d64beba01ee62"}
{"id": "train_039599", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["governance", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "24cd7ad76a0621b6a5f91e8446e12267593d04db"}
{"id": "train_039600", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "64a655ffbf298d843456b63f7bf1c3216d057de9"}
{"id": "train_039601", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "fa25138b6634b642946aa46eeded22dcf6b4f86f"}
{"id": "train_039602", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "35e721b8f396eff32250a277c8983ed2315e629b"}
{"id": "train_039603", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "cd8f0496f945ec8521eb97c93b993915140fd921"}
{"id": "train_039604", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "84474590857b165e22dd1d76d4ed12f3b2e6557d"}
{"id": "train_039605", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "75c79b299ca5fafa81b06f0f92aa81a0635f22c3"}
{"id": "train_039606", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["coding", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "7b9f35dd94e5d077cce6798f2d557d111affe23c"}
{"id": "train_039607", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "f728b449eaf2056c2c29bd0baeb2e82bbcdc33dc"}
{"id": "train_039608", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a628f09532b76a27545a7686e076acca0b23e709"}
{"id": "train_039609", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "78e15df2b90989489cbee52809e8a2cdfe7fa916"}
{"id": "train_039610", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "747504e1b5e4e97ade8da8b3133c37d92004abbb"}
{"id": "train_039611", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "3907e057f69d47874b47e92f64c06a3787262f9b"}
{"id": "train_039612", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "8269f02a66cddbe595d1f348a86ba1d171a97460"}
{"id": "train_039613", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "5d1b0d3a3b357b58ac8f8d9b94f3f7966f5e2aea"}
{"id": "train_039614", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "1e140f7a514fa98ccb1bdf60460720813b5c135a"}
{"id": "train_039615", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e0da0afda80a1706fd8e38f8daa6d90c570b42af"}
{"id": "train_039616", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "52002fa61e7077e82f050aacb8d9b2946a102f91"}
{"id": "train_039617", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "291012d49549693426a9971d2e32c0b98ac0da29"}
{"id": "train_039618", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "53f7ecb79abcf4dc819c461a06c8a35b7404c564"}
{"id": "train_039619", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["research", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "b2a018957eaac0db5e03118fad2fa6a644fa1155"}
{"id": "train_039620", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["coding", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "d4b7a803ec69640ea6c6723469874fc66985b0e1"}
{"id": "train_039621", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "3bada638ba3e4bcb5bcc00684e748573ccddf1fa"}
{"id": "train_039622", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["governance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "84eab9d1b7e53f96fea67dc002ebc3ec85faa1d8"}
{"id": "train_039623", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "b6b250ad6a26846d3e3ac50e885fc0b786600855"}
{"id": "train_039624", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "7920f9091e787a78452df30eacb16291244796a2"}
{"id": "train_039625", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "cf1a2404771bac3134ffe20926cdf6f59e3fde89"}
{"id": "train_039626", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "2a7dc411f37f3cabe98e65c1dded8c221c3d4f22"}
{"id": "train_039627", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3890b91520a52d4cb2584eab9859467588058b34"}
{"id": "train_039628", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "2b7c312b5e876431a8e7416afc40e0fad1fde5b5"}
{"id": "train_039629", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ff2a7ae8eb7e836dc1a469819d7d40bfd7f0bcf0"}
{"id": "train_039630", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "01fd8a1615ad51967aefd670aa6ee5592f46b1c3"}
{"id": "train_039631", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["research", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ca4657b9ea66bda02f712529e5474bf83624f2a3"}
{"id": "train_039632", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "b04d7fadbf010a0774e5097c07251a7e7f14aad3"}
{"id": "train_039633", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e7f38948b839bd7b2eb3d91ec0c9398510413f91"}
{"id": "train_039634", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "3e1daa983aca096c029f8ed3aa09e02f2cb4e4ed"}
{"id": "train_039635", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "5ebeac9f98cb0e3cbb60963cab03e00bd4dfacd0"}
{"id": "train_039636", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "d9572af470a63422049e8b0168e8cafd8d3f85eb"}
{"id": "train_039637", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Java", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e999f01ccbadb579207061a3cdb461f85f139d91"}
{"id": "train_039638", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "85c4c73491d08f834adf309515375e203e83b3fa"}
{"id": "train_039639", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "d33068114ba8b1d9e19781d5f98a6e477e50c025"}
{"id": "train_039640", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "102315d009e93378d87a7cb5a4077f8432290f95"}
{"id": "train_039641", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "168b97b4851f03804aeff67bb1c47500ed537c72"}
{"id": "train_039642", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "5e70966b08f0f1f65ee946bf216b91862583ba64"}
{"id": "train_039643", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "89b409ac29449fd6cfc4ff8f65bac4425e333ce3"}
{"id": "train_039644", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "364fc7ec4dc7ed534216f8a77d09b210e577f155"}
{"id": "train_039645", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "68329995f7ba441a44c575d876bebc794ec6edf2"}
{"id": "train_039646", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "ccca8721c6f91e3d9cc9d8f55e2bcd36647a61f4"}
{"id": "train_039647", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "0d212f38df7fe9d9b3498b98da710201b05c34d2"}
{"id": "train_039648", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "67bfeb5bf076ca579530245505e80a1d3469ef92"}
{"id": "train_039649", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "45fad5d69c681df99342f89dd1cbf9c5208608cb"}
{"id": "train_039650", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "74b0d5e397d5464da1fb2244539aa0c9e3f3d6c5"}
{"id": "train_039651", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["research", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f7305f02c60485fd549d1be8c0660c451aa1b414"}
{"id": "train_039652", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["performance", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "1c1024efbc7bfd5bf9e9c26da31069b68a9dd8a1"}
{"id": "train_039653", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["data", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "93b341d3ea3406b577868f7f474299562e9dbfa6"}
{"id": "train_039654", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "32725e10d52d1229cc742d53fe632a2f9ff91c4c"}
{"id": "train_039655", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "de50c25861139dedf3ab10433c9e42d869f99b5d"}
{"id": "train_039656", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "438dabaa2dde99ab9225ac826816d34955ea6099"}
{"id": "train_039657", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "7384c28528ef216ecceacb17d8d9c160014fd5c2"}
{"id": "train_039658", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e57ca0689b8494c2120d12e529c07a2595e541ef"}
{"id": "train_039659", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "752cee9c668c2d6d27c14fa611984f41ee889e43"}
{"id": "train_039660", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["coding", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "bbbc1241dacdeffc06b77fc010200163d9fe16a0"}
{"id": "train_039661", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "3505aa8c8cd524072bce62d9134eafa4f9fea911"}
{"id": "train_039662", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "86ca7dab7f6f5304c8b0029f27d9d3c57c315e7b"}
{"id": "train_039663", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "b1f54ef5d390a347ae746df4f7b9c8c762c0b418"}
{"id": "train_039664", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "9c11c9e5cb464f96222e0c583aa9fba9f85db801"}
{"id": "train_039665", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "02253f77f6c5abbb1fc7166ca98d7b0473ff9e9e"}
{"id": "train_039666", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "c55b98db57ce80f0c222653e11ae389ff81d8c14"}
{"id": "train_039667", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "089a8b99712ad65e1db4b5037ea607979c56c3aa"}
{"id": "train_039668", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["research", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "e4b1f098e79f15c58cb530b74d476860d435e0f5"}
{"id": "train_039669", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["evaluation", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "023a8924ab224fd020fd893fb0ee7fa7e1ea2e91"}
{"id": "train_039670", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["research", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a2c5e24b05323534dc840435c73a96c3d04756bc"}
{"id": "train_039671", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["performance", "evaluation"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "6e24dee98e1af62dc9ee8b1f8bf41a078a889661"}
{"id": "train_039672", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "1b97eb963f1e8ee0072b68a8d1d765721804590e"}
{"id": "train_039673", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "5049860b1d7aa1795d0410315822b57adab9cab3"}
{"id": "train_039674", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f8b398ec268dfda168f107cef26edbeb6abe49e7"}
{"id": "train_039675", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["data", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "4e6b2197ef70af333afcf18dcf284f6f100ea01d"}
{"id": "train_039676", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "f2b6bcf6d33c080496d5d750a72e8cd581baa786"}
{"id": "train_039677", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d6261ae344b31591e7f2b8c6410068239e3c2cfa"}
{"id": "train_039678", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "63cfbcc751193d1d7e04dc6b7bc351de9cfe62f8"}
{"id": "train_039679", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "10ac123abeb1675b5c112e98138d5e7c98596484"}
{"id": "train_039680", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "b3f2270c166060f36333c70c34488665fb312e67"}
{"id": "train_039681", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["agent", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "9a87e10808bd57af59642c9f33e3988e14fed184"}
{"id": "train_039682", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "182b6b0ea544b4ef3ed9afbecf916c3aa8b0d37f"}
{"id": "train_039683", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "078d988d77765103385f65cfbeddabff2871da88"}
{"id": "train_039684", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "41791728e1db3b6ab26e58d39eff9c9b41d8a0c9"}
{"id": "train_039685", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c1a6776d212e63f37f02443f57ca56d707946987"}
{"id": "train_039686", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "8e1f48cbd64f4bbf14e7dc1e7b75185acd281064"}
{"id": "train_039687", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "703a109d0a6e00fcb3a5f4e3731fa00d5a03ffce"}
{"id": "train_039688", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f17fa60bf2215e0ed155132abffbb81cc7f41c7f"}
{"id": "train_039689", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "e9dce1ebbab75b51a441d58eb4c59eabf0cdac0c"}
{"id": "train_039690", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "79679a2565d38789e0e7452078cb4e8ec34458a5"}
{"id": "train_039691", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "cccc0ca89821cdbe01dc4bdc8701bcbaa505c765"}
{"id": "train_039692", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c9727fd78b95c718645f7f67be17134f9f5e55e5"}
{"id": "train_039693", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "76fedf100e672d216f60656cfacd06e6726b5d45"}
{"id": "train_039694", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "3bff15f03f20202bdb26efdc979cbbe51bb1b3e7"}
{"id": "train_039695", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "79bf2825d70f3baa185fab334658ecb47805b6ae"}
{"id": "train_039696", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "22f89c2ce6b726398ff16721973f60491ca1fc18"}
{"id": "train_039697", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "15b24b963651b571ffa56ae3d2e93a209b85a1de"}
{"id": "train_039698", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "542d97acc90e4ab4f38c6633a5e91b34e7da8970"}
{"id": "train_039699", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "a6c6f95af1df7222b17dc50daf8dd8a675ee5d0f"}
{"id": "train_039700", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "df3cd0bf14f7255139b07fb89beb35a3a8f4f300"}
{"id": "train_039701", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["research", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "a59df88bedd438e17359c75f157799522197669c"}
{"id": "train_039702", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "587dc91e562cc52101fc693e6b23a2aacb676481"}
{"id": "train_039703", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "5a9db075a1627acf0a0a5820469452485915484e"}
{"id": "train_039704", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "9bc9131ff5994ecfef23f3de5b322289eebbf588"}
{"id": "train_039705", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "d908b9bd6c1d18c929657c909297102b5880d34d"}
{"id": "train_039706", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "6e701e10f854bcb6041d25dd13c3a09c697220fb"}
{"id": "train_039707", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "01816712960cfa579f831f8f2f1f20b3b82deea3"}
{"id": "train_039708", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "32eae54258de61f8a51795c2acf742934995d853"}
{"id": "train_039709", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "2a1828d9fa416f7ad9fba142adfe6f3d97e2993c"}
{"id": "train_039710", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["governance", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "40660011115e1fb1da46834f611020d3f7b65be0"}
{"id": "train_039711", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "390f6dd48612938dcd9bf831b05401bd6e6814e0"}
{"id": "train_039712", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "0fc480c5496807e8fabc9a6a3242c15a71a2b712"}
{"id": "train_039713", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["data", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ab5b97d785c4af6715a9b76adbea3b51c9914195"}
{"id": "train_039714", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "C#", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "e1cfd737cc40c460ba3fb1b145d47eda11950297"}
{"id": "train_039715", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["agent", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "1257341cbe26cdf5a9bb74498477edadac1778e6"}
{"id": "train_039716", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "4b969ed6a585f56931d21a5ce0f0919cda37d95c"}
{"id": "train_039717", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["research", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "26a43579ba358cc2760ce64271ebdc4716479cc4"}
{"id": "train_039718", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "483458f4446647f7614f9ad3cb4bc6b282c05f30"}
{"id": "train_039719", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "43b72f624df4ce0ce7e0291d2390405f9a3202ad"}
{"id": "train_039720", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "91b51738183f1022d4dd902317752a9179966705"}
{"id": "train_039721", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "bddfb5559431d0e9e2c0eec66804dd06d156a155"}
{"id": "train_039722", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["performance", "security"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "a0ea08be45e61915669f9865d5799986cf9d829a"}
{"id": "train_039723", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0792983c1c4bafcf36bd9f2e0c4f3fd21c27c677"}
{"id": "train_039724", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "63e83675f11b3adb974fbff642072766b64bb4b7"}
{"id": "train_039725", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "943aaba7a371c1fdb692fb087bde346f888145ac"}
{"id": "train_039726", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "512adf868d1a1bf17e877a45fa59d04d15770fa9"}
{"id": "train_039727", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "92bb03900a7e2eb0c906682b909bc978b1fca6f7"}
{"id": "train_039728", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "75c27a07e27150d70ae16a71824e54765f04e373"}
{"id": "train_039729", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d27bac1d610f601b046157664d75b30ec4259860"}
{"id": "train_039730", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "28272db1a70b1ff616561e4dee0fac2756d55736"}
{"id": "train_039731", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c293bd74e67988a00552c4c69b958fae6c283545"}
{"id": "train_039732", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "2ec48ad22ad7e488a093ce5f2f9ad64ac5a6bc01"}
{"id": "train_039733", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "877e8e3f36ad9738d9186b3051cfeaeebd55b275"}
{"id": "train_039734", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Rust", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "eedd59b9ef2f39029374f56541b177c132f4942f"}
{"id": "train_039735", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "33d5f9408e54afcab24561ab7f3366303e0f03a2"}
{"id": "train_039736", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "3ac46d4bb6597f9dfade8252f1df98c4d367e336"}
{"id": "train_039737", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "fa47b86a65c092b0b34cdd7ea51ea333d7e280d6"}
{"id": "train_039738", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "308be70043dd953a0056cfcf82f2ae5977c88ebd"}
{"id": "train_039739", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "2aa938ff8f81d7bed2005e17f91f6e92cbae1516"}
{"id": "train_039740", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e8d4f6d634de0122d62d80f97e19b9f13345c564"}
{"id": "train_039741", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "bd7362e3b23109bc0ef76fe2c8b6d4b46d393b8f"}
{"id": "train_039742", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "3b64cf98d90bacefca48ae9c7b4cd9ee1cbdcf21"}
{"id": "train_039743", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "1ca4cdc8f6c576cc90db5e5ac5cf11f1f66ed530"}
{"id": "train_039744", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "22e2a52c51b86a5c9ca5eb16f1941291544a7c6c"}
{"id": "train_039745", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "aaf2bb1bd6678773333f842ebfd34d851a3ad47e"}
{"id": "train_039746", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "499673cf76040af3879cc8387f765a0c9253807f"}
{"id": "train_039747", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6ee8393a595719d67f1be4dd3bb33decdb9fcd48"}
{"id": "train_039748", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["security", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "a8e968982663fcbf14699b954df55103d52c3f69"}
{"id": "train_039749", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "9516a6c014d8ce61b308abc65416c4c26f07b732"}
{"id": "train_039750", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "8c5b1bc2fbcdfc51dcea89d0818423d4bd89775c"}
{"id": "train_039751", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "3316b54f5280a2a35117814980051840b0a09765"}
{"id": "train_039752", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b3ebec52565684da49e2fb9b618e662aefecb8b6"}
{"id": "train_039753", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "192a371896d1ef9df1f36a904b1589430c941e6e"}
{"id": "train_039754", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "63f5fc8e110c2651d97ad904b58c1d07aa32e574"}
{"id": "train_039755", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a0f7b12259571919b69c6b1a72cb016ce82e62cf"}
{"id": "train_039756", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0f023b9ea076ef899af8b5a47ae947a2122e6c78"}
{"id": "train_039757", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c18f0b4342f943ff0671c897f82639f84193631c"}
{"id": "train_039758", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4e0224010a81f8a459272b3153fb1718a463fa85"}
{"id": "train_039759", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ff7e221ad09aa4d39d045eda453598af05b55037"}
{"id": "train_039760", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["research", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4b790bc69249f441994f9c7d4246f9c23d07dcbf"}
{"id": "train_039761", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["evaluation", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "25dbe206cd1ff608c33ecf563e7f0783d50f7c3f"}
{"id": "train_039762", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "9bcb9a8adb7cb2283e0f2f47515584507146b317"}
{"id": "train_039763", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "300f9d1bc5865fd3561d0f513a306c925bdd27e9"}
{"id": "train_039764", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "c8cbb6bb53a7e5a68f3ca439e97bfa4f7da57955"}
{"id": "train_039765", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "710479b385359edd4a1140097b03cb115ba77551"}
{"id": "train_039766", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "7faf35083e15079b0054a471b1dffb3474bd549b"}
{"id": "train_039767", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "945e8dc7e7b6e4b2bc772de6c2b3d2c21939f136"}
{"id": "train_039768", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "8ba7289e014d5ddf156a2b96d4992eeeeabd3ad0"}
{"id": "train_039769", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["governance", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "950a126f67943efaa77c4f34a4cc60aa4d4d80c8"}
{"id": "train_039770", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e8bd291339956c936bce6e95c646592253a861cf"}
{"id": "train_039771", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "agent"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "a3a9d6b515d8c0d251e5b4fc037f2df5cc96e43d"}
{"id": "train_039772", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "4dd0109fe6d2940e1f03d29af2ebb0ac2b04e999"}
{"id": "train_039773", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "f7c1f72d1aaf3332785f535ad9ff73e2e64e3e89"}
{"id": "train_039774", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "coding"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "bcd42e0ee074bc9b3489aa3b9831b92e6471e8f1"}
{"id": "train_039775", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "cd0057b81c844d9de517c70fb7b20743d1138e56"}
{"id": "train_039776", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "7fe2c217a2b88bdf10c30a418b2540e50ecef71f"}
{"id": "train_039777", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["agent", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "832a58e63a378528f9f6af2e326a018e8684c547"}
{"id": "train_039778", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f5575fc6818ac86d66168760f255454f5f163fcc"}
{"id": "train_039779", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["evaluation", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "75a06fac207552ae44e2062e80774c79fa3937da"}
{"id": "train_039780", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "800844eac4a1740202ccb7209982142d3a9d9c24"}
{"id": "train_039781", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["coding", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "e313c012d78efdbbb4b1b8a2df9569c86a02cd67"}
{"id": "train_039782", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "4dc4800412318dff6bc3ac61233e4d79a8c0e490"}
{"id": "train_039783", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "2059e5c1317333502c960a79117e7ca8a46308aa"}
{"id": "train_039784", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "c9d6dd78e2500010a6871c2da8adbf3cdca2908c"}
{"id": "train_039785", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "3330b0c1f076dd6a55402677fffdc4d84a89a6b8"}
{"id": "train_039786", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "4178464ee52a77f5d904e7bd478bde6f20f5f649"}
{"id": "train_039787", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "a628f09532b76a27545a7686e076acca0b23e709"}
{"id": "train_039788", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["research", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "91791f4f6ac7b25fbd9aa9ffacd97c5e7d982efc"}
{"id": "train_039789", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "b5911f1221a78f93d3f72a38150082e9b79ead07"}
{"id": "train_039790", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "7bf869c7a5cc71e6a6829058c9fd1a29623205fc"}
{"id": "train_039791", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "cdaa503bcbec9e840582e1394e0c593e6d3c7ff3"}
{"id": "train_039792", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "1bb8d723c0df1c9ee401b31571c72725b2ea72cb"}
{"id": "train_039793", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "979ac33282efde343460ec6295a0d17b2d2181c5"}
{"id": "train_039794", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "03453660e86fc94bb655c77b7e22ab54a0986cf8"}
{"id": "train_039795", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "69a183765015a600a675f0a689738b3aa56ce51f"}
{"id": "train_039796", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "26af262c46fe82c80a7e85a125d68726d3cdd427"}
{"id": "train_039797", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "c16db4dd542d681ed9fc055d74f8c8d2492424e5"}
{"id": "train_039798", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "a6320da484bf4dd26b5f656052dd8449a854e04f"}
{"id": "train_039799", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "8d8bc925cf8204ea011da25b605c0137987a4aac"}
{"id": "train_039800", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "342aa9450a640d8a9d813ac0a65bfcdf33e4db1a"}
{"id": "train_039801", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["coding", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "e8863e49f4c7666692ac5e726f97ae862135af71"}
{"id": "train_039802", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "SQL", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "0d7601834335080442244e7a2dcb307f4a6591de"}
{"id": "train_039803", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "5a2c836c28ceb066996f670f8dd76ea42459937a"}
{"id": "train_039804", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "9efd66dd3a33013642e19efba92a50fbad0eaee6"}
{"id": "train_039805", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "12b875663e2f5b51f7b3541549127c113257a919"}
{"id": "train_039806", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Bash", "moe_experts": ["data", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0e02ad4af959338807dc92ca6cb620c2ba071605"}
{"id": "train_039807", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "a7a8c3797a8aaa734d0d13ada49fab53c1d044fd"}
{"id": "train_039808", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0e940b30ec472c8bd2455ea96f1107e4264810e2"}
{"id": "train_039809", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["performance", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "26f4b436a972a6a37d2026cd40addc4e42c967e4"}
{"id": "train_039810", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "2bcbd4d69d306720e17139f82efac1997dfbf8e7"}
{"id": "train_039811", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "aca90055463714a920406d9c8ee0f7c021a7f0d2"}
{"id": "train_039812", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "94ceffaa7f75921e7409206acdc15a7d25c933e3"}
{"id": "train_039813", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "a7bd5d56ae09c512ce0cb36a3d32bec065798d81"}
{"id": "train_039814", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["research", "agent"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "3113d90880d21c88698006f5162adc59a30c8e73"}
{"id": "train_039815", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["research", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "9b355aa04352046e3a13e1043bf5f0cab2685788"}
{"id": "train_039816", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "9d179aac646a1d9fb5d7698b661a30b64d7c44f6"}
{"id": "train_039817", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Python", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "d597258fa4d26bcc2bf47e33c63b33e4198fcfef"}
{"id": "train_039818", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "54e78e90ad14f9a9f570670910416d8ab5619a79"}
{"id": "train_039819", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "beabb263c9bf6ada4a73be63f786d82557f96ca2"}
{"id": "train_039820", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "ca980fc195d0ea84e1bed2aee74eb5c597f7f69a"}
{"id": "train_039821", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "c14c4d27b7db197aae041ff26d6e7531a2c6ba38"}
{"id": "train_039822", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "24cfc09f00ba3e9600765a1035b114b2ae79a319"}
{"id": "train_039823", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Python", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6317d753841faa842b3f82fe907a17c6eb48da02"}
{"id": "train_039824", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "374aab7b53f099ba7ba0de0e81d2dfdc71b908b9"}
{"id": "train_039825", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "6e0b74481a44b0379d791cfb05dc5ea85ef9019e"}
{"id": "train_039826", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "48613139ed9ba9f4d89947f04eae0d9439597d24"}
{"id": "train_039827", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "f0d45a0d7c221c0dc2ae281b11b7416a4430db79"}
{"id": "train_039828", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 10,\n  \"confidence\": 1.0,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["security", "agent"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "ca6d49e1630a2caa993fe1910cc0e21744fbb74e"}
{"id": "train_039829", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "d4ccac02371e0deb6a7df4db9248d2a93b7f9451"}
{"id": "train_039830", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "c6027552d83858dd21b5e01500ef4ab7b5ce5635"}
{"id": "train_039831", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "b898bad69ee8e87d746f9da4dccb00159558caa2"}
{"id": "train_039832", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "121ddfd9b0f3a81e432f834a8c852a2333a6cb93"}
{"id": "train_039833", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0cbd10495ca9a2d6347014ac99654192d0e10f12"}
{"id": "train_039834", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "9f2ac2dd49c9b6f16c129aff9d0fde4fcf888c2e"}
{"id": "train_039835", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["data", "performance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "75eca4314eb5fabc5266711fbc534c6f6106ddb1"}
{"id": "train_039836", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["coding", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "7802b36cf82193e4ff10a50323664448c5c25b9b"}
{"id": "train_039837", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "a2d0565068fa1cc53d755abea691ce478c176664"}
{"id": "train_039838", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ee47fb252944a86e6fac0ffcb441b4ae2027cffc"}
{"id": "train_039839", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "191ad986b831162f1360b4c4d90082e57f2bf8bf"}
{"id": "train_039840", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "coding"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "afa6e9a5970d52d1b1b636ef9f6af777dee133f4"}
{"id": "train_039841", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "data"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "ffa6f9ba3819d99b7a707b743d20967af7c0e773"}
{"id": "train_039842", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "6a88082401a612ba4382dd372dd66b825ef11abe"}
{"id": "train_039843", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "68b154edde897c9b4d5efc7803846ae06f31e618"}
{"id": "train_039844", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "e3563f395dfe2ffa13ff30d6ab3d155e20ea2599"}
{"id": "train_039845", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7a40e01ad299946b380c47fde21670d4a9a3363d"}
{"id": "train_039846", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "4de414589c88046f4e9d8a923825021e2fcc50a7"}
{"id": "train_039847", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "68fc7531bace5be46990d38bf062b9e2421a4883"}
{"id": "train_039848", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "4f55f4d876656593445686ecca258edab0a08817"}
{"id": "train_039849", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Java\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Java", "moe_experts": ["performance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "7363dce65639c3b3ecf8353c2c7aaeb5a4b8e0fb"}
{"id": "train_039850", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "3d7320aef923c10f9d26af8b40e5b500728d1ff9"}
{"id": "train_039851", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "78239f9c1fc7839798e5fd2a814609e3061e38e1"}
{"id": "train_039852", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "3dc50d2b883a431445c11596cc38c53361211054"}
{"id": "train_039853", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "governance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "eb2503d55cbe812fbc5e304c4d427087dff9aede"}
{"id": "train_039854", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "027d82e9a3e537bf91b189700ce318cf9d7b206f"}
{"id": "train_039855", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "03ddeb8fd10d781abcfce77622abeb3c53ef0858"}
{"id": "train_039856", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "performance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "845809eeb05420cc6d6831cffbcc074d1d99394b"}
{"id": "train_039857", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["security", "agent"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "c596d29d903966f48bf418180b967e9a070798d6"}
{"id": "train_039858", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["security", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "d008029c3500353b1bcb09fdee1390deafc1c1b0"}
{"id": "train_039859", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "59259cc20d7f7e9a5c3472332397eab23d6a38d1"}
{"id": "train_039860", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "edde062fe33371071b8ec5fa871ce2cc6d40ae7a"}
{"id": "train_039861", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "960482ff383554339b76c91e571996e0a97e3ef6"}
{"id": "train_039862", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "b78206c6f53cd8ba64c186c8178806d115dcbce6"}
{"id": "train_039863", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "dd176b5be9498fd1436d2d942a0c72c9b36b3f13"}
{"id": "train_039864", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "data"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "ca0c8475da60b1db08044075976dd9b38228ce0a"}
{"id": "train_039865", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "e8f72d24199833e951359b8ee5a65bddaca79b1c"}
{"id": "train_039866", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["data", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f134244e21860071018f452390be6bd83b9fd644"}
{"id": "train_039867", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e906c11f7d92852bf0899926dc6116147c17422b"}
{"id": "train_039868", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "JavaScript", "moe_experts": ["performance", "security"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "8180bde714b32f871e04342d4023198a8cf20a09"}
{"id": "train_039869", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c8f58d9c1e9a7c91a20cbd02b1cdd217e91f9f17"}
{"id": "train_039870", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["research", "data"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "1faf6254e00a4eca538122ecbf563bac31b4d606"}
{"id": "train_039871", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "governance"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "52bc9ecbd01a8c04c67ace07a85d7b496436c810"}
{"id": "train_039872", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "79274bed166f7221c6c3f986ed0ec02539fc5fb9"}
{"id": "train_039873", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: Go\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "7b1e9b3775af3cc7b96dd9cb5a15ffba35ae824f"}
{"id": "train_039874", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "4982a5c546747167171e6cac20a0485a6acfad00"}
{"id": "train_039875", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: advanced\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "security"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "2ab17f50df5ee830d501d849dd9d3f6f43c3a24d"}
{"id": "train_039876", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["evaluation", "performance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "0d327cfa51a5b2c61da65cf162e6e2ed40f5a635"}
{"id": "train_039877", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Extended context, memory, and repo graphs\nDifficulty: expert\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 8,\n  \"confidence\": 0.8,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["security", "data"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "9b362388f5ad1eda238f17a2a993c07def0f3aed"}
{"id": "train_039878", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["governance", "evaluation"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "01a839f77a203b846f2aee9558bb7f6d0961ac00"}
{"id": "train_039879", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Java", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "0bf03ff7b0c1d614c506f20a10bbfe1e3aeb1e91"}
{"id": "train_039880", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["performance", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "794cde529d22924f68c52e6acfa9dcd40648690e"}
{"id": "train_039881", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "934c99875848e8f328aa75c0a32fce692821f863"}
{"id": "train_039882", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "agent"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "9eca8a806548a838b75133a4898a28014718c96a"}
{"id": "train_039883", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "security"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "ad4849b876e119ebbfaf8dffcab9b6bac394090c"}
{"id": "train_039884", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["evaluation", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "0fe37ba9c09a0a8059494f877f395ada7e478dee"}
{"id": "train_039885", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "2a4c28f357742b6b873a554a993399be62ff68eb"}
{"id": "train_039886", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Python", "moe_experts": ["security", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ca5c4d9e82049abd9f4698ba3ffd12f4cf81ad34"}
{"id": "train_039887", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["research", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "fd785beb056071141ef9f49c879764a30436a2e1"}
{"id": "train_039888", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Bash", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "6f1ad019038204c5838f7b7837c277a4cd118eee"}
{"id": "train_039889", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "94eaf3604b600e490bb20fc7f8cbc3d371f8673d"}
{"id": "train_039890", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "agent"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "e59680cf1f12e89264047abdcb79bc7de45d050d"}
{"id": "train_039891", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "SQL", "moe_experts": ["coding", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "59bd73cf5f39521d3b0540b05c12f6315cd613d9"}
{"id": "train_039892", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "299282ae16ae2dffaa87c4bc34b4080d967fb3c4"}
{"id": "train_039893", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "f5d0fd588dcfd2bcab65e4b8c474e419d21d0242"}
{"id": "train_039894", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["agent", "coding"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "62392879e11868588e97133448a53814456314f6"}
{"id": "train_039895", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["security", "evaluation"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "1c30db7f22b844be2b050ebd2982477c9f486a9d"}
{"id": "train_039896", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["security", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "ce9f4ed6fb088b9757e5d9a6a5f3889bf7ae0ac9"}
{"id": "train_039897", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["agent", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "b682703ac19ca74c9009eac8a2e05731d418b111"}
{"id": "train_039898", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "c9b04588e5084c67e0308685c80d68370e8de496"}
{"id": "train_039899", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "deeb619615ae6b0a05651542dfe5b62520e5a2b7"}
{"id": "train_039900", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "TypeScript", "moe_experts": ["coding", "research"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "5c8aa178d773f3c42b3c9942d4e09d0cb442893a"}
{"id": "train_039901", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "c2b6efefe1e24aab94f959f966f306c57971a661"}
{"id": "train_039902", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Rust\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "data"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "791e7f72f39fddee7bd68a28223ecb15cf25cf79"}
{"id": "train_039903", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["performance", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e089ea2c453f800a7e2a738b36421fa22051b576"}
{"id": "train_039904", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "0909b54e25e9724de69788e37130d757d01bdac4"}
{"id": "train_039905", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "C#", "moe_experts": ["performance", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "8bd99303ee0d854a661a9894ab9b05ebbec2989a"}
{"id": "train_039906", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "9d2eda413bcada2da67c3ad8b76eb9c07f235fbc"}
{"id": "train_039907", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["evaluation", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "b58a2cb496e0b4aaf1072a76baf60c3dae81c2fe"}
{"id": "train_039908", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "bc5fa6fe3bd46f49cc9209b95e977a8909d34393"}
{"id": "train_039909", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "94b6b014cd6cd93d7d82462bf481f4bd87958c9b"}
{"id": "train_039910", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "8f0ce6db3cba4e8882d8e0ed594b028bb7c8b139"}
{"id": "train_039911", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "53f20d336374bad1d86e6db899a9c0eaf7ccce5b"}
{"id": "train_039912", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "0e3571708058756fc79ee8a30370639c0adc5c4c"}
{"id": "train_039913", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "JavaScript", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "28d4ad5b8ddd9cbeaab38fc8dd09a2feaa065ad3"}
{"id": "train_039914", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Java", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "6457ae641b66ed2bca0fc03321b007a51e59797f"}
{"id": "train_039915", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["data", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "52ee8812434424a43de894bbb892aadf5fea7e02"}
{"id": "train_039916", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Latency, cost, throughput optimization\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "7ff531d476b03b47f275d5656031e10fc5b1b6d2"}
{"id": "train_039917", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["agent", "governance"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "6e5b0add9ce4f00844bab6b15cc9126fa95ea84c"}
{"id": "train_039918", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "3f4ecdee8325fef4f438b3d740f3de3bd7a0050d"}
{"id": "train_039919", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "c8e47811851385fd42391734077130169b1dd373"}
{"id": "train_039920", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["agent", "security"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "680ba1b60a2bc8a764056273849b0e7f94c40db2"}
{"id": "train_039921", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["data", "governance"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "91d503a7b28a21c3b2509d50345f5ce8b6bd0611"}
{"id": "train_039922", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and scaling laws\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "performance"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "6a22999336b64388358e41faa20b0e78d2646d7a"}
{"id": "train_039923", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "8c9011817cb8e4016db6b1ea63b5ce2b2eee7dcf"}
{"id": "train_039924", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "94b6b014cd6cd93d7d82462bf481f4bd87958c9b"}
{"id": "train_039925", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: TypeScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "TypeScript", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "e1ab0cd71d20280fbfb7659ce9d52edf93fd6390"}
{"id": "train_039926", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["agent", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "eab556d42943bed383a13b09781870200bfd5301"}
{"id": "train_039927", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Rust", "moe_experts": ["security", "research"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "004bb1e27cafed9eb0abba609beff2f2fe64d68f"}
{"id": "train_039928", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "a45a39c6666714236fe1a1379a7658132ea4b723"}
{"id": "train_039929", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["coding", "research"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "29e19804b05f616f785d07e17f13af20b468a384"}
{"id": "train_039930", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["security", "agent"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "b1f957dc554a68cdefbc5ae00ee5be2be3fe7835"}
{"id": "train_039931", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["governance", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "1ac7733071ff97ba3d2e5656e02eb4103249a9a6"}
{"id": "train_039932", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "31c18fef10b18d3cb5838999ed873cd85ae5bd32"}
{"id": "train_039933", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: Bash\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "facc31fab453a9380b66f6d5e9287753e4968117"}
{"id": "train_039934", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["performance", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "a75f4854534bd9eb911335ca927fa1768c8fe22a"}
{"id": "train_039935", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "e2fe8a0916909b7510bcaa86aedd73da183c27dd"}
{"id": "train_039936", "created": "2026-01-01", "topic": "Code-specialized model families and scaling laws", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Code-specialized model families and scaling laws\nDifficulty: advanced\nLanguage: Java\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Java", "moe_experts": ["governance", "research"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "4fd77079a6337364ba3053a6bac69687b28127b4"}
{"id": "train_039937", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "governance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "5694651f444b66fd8f1dc2592d7f5940e58bdd6f"}
{"id": "train_039938", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["governance", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "587dc91e562cc52101fc693e6b23a2aacb676481"}
{"id": "train_039939", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["coding", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "a9be2e25a1c22e35ef5ce123d22b386a14d14769"}
{"id": "train_039940", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: SQL\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "72971499ffcdf30cde9eea808f7c81083f27c1e4"}
{"id": "train_039941", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "a5e0d0627820de9cf856b6fd1738a853e726852d"}
{"id": "train_039942", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Bash\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["security", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "d31d1b7a63f1c4bcc64ee2a4d8162a6938473a77"}
{"id": "train_039943", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["data", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "4a6203341199981659b24245a26f24f15d89d53c"}
{"id": "train_039944", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "coding"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "55ba7cb1bbf697f52807962c0df8ad0fae1d6002"}
{"id": "train_039945", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: JavaScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "JavaScript", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "42e60aa46767e6722eb85886eed2870b6ca191e7"}
{"id": "train_039946", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "governance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ad152427d1b71a38dedf9e966d9388d2124033e3"}
{"id": "train_039947", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "614662976a4d4d8333b340b4b38a5f94ffec3651"}
{"id": "train_039948", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "0bade7dc740a0926c1d3191f2081f9ded861ca81"}
{"id": "train_039949", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Python\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["coding", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "9c6446545f21f0ee0c952470d6f4c8ba9c7831b9"}
{"id": "train_039950", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["data", "governance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "beaf43887e25aa37c2521740b5a59d3b81dfcea4"}
{"id": "train_039951", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nCode:\n```python\ndef agent_loop(plan, edit, test, issue, budget=4):\n    cost = 0\n    for _ in range(budget):\n        p = plan(issue)\n        patch = edit(issue, p)\n        ok, _ = test(patch)\n        cost += 1\n        if ok:\n            return patch, cost\n    raise RuntimeError(\"Budget exhausted\")\n\n```", "metadata": {"language": "Python", "moe_experts": ["performance", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "803c2f912daf45686171e13ae54d45e30676c42c"}
{"id": "train_039952", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true}, "hash": "ebcc4227996926853a7f878b4554a5962ff32d76"}
{"id": "train_039953", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "performance"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "f334495ccc8a732ba320c4bb8cfa9bd0dde56663"}
{"id": "train_039954", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "SQL", "moe_experts": ["agent", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "2c12a9571dbeb2ef033bd9066d2f56acd036abbf"}
{"id": "train_039955", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Bash", "moe_experts": ["research", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "50ef229b0bbdad1f0f760eec1981b1866aec73cc"}
{"id": "train_039956", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Python", "moe_experts": ["security", "evaluation"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "2d2717510e415ff658e124280320b2cf4d6181a6"}
{"id": "train_039957", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "7aee87d911789afe1cdbcdc043539d165ef2260d"}
{"id": "train_039958", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: Java\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "4a261f3840d93992880f971d100825907fd36097"}
{"id": "train_039959", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Failure engineering and resilience testing\nDifficulty: expert\nLanguage: Python\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["governance", "performance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "3c3b89e9be57a31d42372f0a9996afe792bd765d"}
{"id": "train_039960", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["evaluation", "coding"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "9dc8a7702e949bcab039f18323ab5bf69b3f9c0b"}
{"id": "train_039961", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) routing and sparsity", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) routing and sparsity\nDifficulty: advanced\nLanguage: Rust\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "23ff9244e2fa6917f453b9599709b6bd958cacaa"}
{"id": "train_039962", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "C#", "moe_experts": ["evaluation", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "ee8372cb9bf108ac99520c29ac6259b0d5e6533f"}
{"id": "train_039963", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: C#\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["performance", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "32ab6fa094f3c8e97417cddaa1401469149c270f"}
{"id": "train_039964", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: Rust\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Rust", "moe_experts": ["governance", "data"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "1257341cbe26cdf5a9bb74498477edadac1778e6"}
{"id": "train_039965", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: TypeScript\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "agent"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "d514e697d93c8ea193b9efca686c608658972cb7"}
{"id": "train_039966", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: JavaScript\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["evaluation", "governance"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "66812bf0ae0dd8a24629595049c3bcadc565db2e"}
{"id": "train_039967", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "data"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "88399b120fe5d35a75af52f82193878dfe08f4fa"}
{"id": "train_039968", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: expert\nLanguage: C#\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 9,\n  \"confidence\": 0.9,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "C#", "moe_experts": ["data", "performance"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "9fae4cfd4e07ba42393e608a910fb2c630965014"}
{"id": "train_039969", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: JavaScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["data", "evaluation"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "d40a5d98e042c1cf83e2c8e2fbdc92d9fa46cf31"}
{"id": "train_039970", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: SQL\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["agent", "evaluation"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "b20da4df96c103e7571c700ad01575060667fc84"}
{"id": "train_039971", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multi-agent coordination and role separation\nDifficulty: advanced\nLanguage: C#\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["data", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "a778921611932b713202311ab0254226840a9a8d"}
{"id": "train_039972", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: expert\nLanguage: JavaScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["security", "coding"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f907d9fa2399b051137a7123f4002dfb9ab24407"}
{"id": "train_039973", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: C#\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "C#", "moe_experts": ["governance", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "7ee6f4c48b7d0e31499ad7b9ce0e1bee642e3e3f"}
{"id": "train_039974", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Python\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["data", "research"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "fc5c00dc8fecbb5f147ff3faf1a1d3e349d336cc"}
{"id": "train_039975", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: Java\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["security", "research"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "cef98ee25a9fc7c40cad115659dd4e4bd29e6661"}
{"id": "train_039976", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Go\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Go", "moe_experts": ["governance", "research"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "dfe5c363a1ba32548eda56111f61c78803f8dffb"}
{"id": "train_039977", "created": "2026-01-01", "topic": "SWE-bench real-repo evaluation and leakage control", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench real-repo evaluation and leakage control\nDifficulty: advanced\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["agent", "data"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "407f3f06711a590aab14a933c288ebcbfef41a12"}
{"id": "train_039978", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "self_grade", "difficulty": "advanced", "instruction": "Task: self_grade\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 7,\n  \"confidence\": 0.7,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "Go", "moe_experts": ["agent", "research"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "8cf920a1f6ee32e8e70df34de68d2bcfd9e5e321"}
{"id": "train_039979", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "SQL", "moe_experts": ["governance", "research"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "1b078be6876edc8122175d84c4970cbf34d13d3d"}
{"id": "train_039980", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: C#\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "C#", "moe_experts": ["research", "security"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "2809d5345ec2ec27cbbc0ce697f0f1bd7f8b98e4"}
{"id": "train_039981", "created": "2026-01-01", "topic": "Dataset curation pipelines and quality attribution", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines and quality attribution\nDifficulty: expert\nLanguage: TypeScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "e0073c18d96168a2c25b4b34112b9374c556a89c"}
{"id": "train_039982", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["agent", "performance"], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true}, "hash": "ac6a0a5ab593a4f24d8c1af61842547815a3992a"}
{"id": "train_039983", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Self-grading and automated model critique\nDifficulty: advanced\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "JavaScript", "moe_experts": ["coding", "evaluation"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "fe2dd8b91b357d619b6ec51fbe9df9745a8ed0a2"}
{"id": "train_039984", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "tool_trace", "difficulty": "advanced", "instruction": "Task: tool_trace\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: advanced\nLanguage: SQL\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "SQL", "moe_experts": ["research", "data"], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true}, "hash": "6ab1f3b1683e4247131b1deec45e136fc7cfb66e"}
{"id": "train_039985", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "tool_trace", "difficulty": "expert", "instruction": "Task: tool_trace\nTopic: Governance, provenance, and audit logging\nDifficulty: expert\nLanguage: Python\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n[\n  {\n    \"tool\": \"repo_checkout\",\n    \"args\": {\n      \"commit\": \"abc123\"\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"unit\"\n    }\n  },\n  {\n    \"tool\": \"apply_patch\",\n    \"args\": {\n      \"files\": 1\n    }\n  },\n  {\n    \"tool\": \"run_tests\",\n    \"args\": {\n      \"suite\": \"full\"\n    }\n  }\n]", "metadata": {"language": "Python", "moe_experts": ["evaluation", "coding"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "ae32b9bae03142547db515d17f6d68aa63cffbeb"}
{"id": "train_039986", "created": "2026-01-01", "topic": "Failure engineering and resilience testing", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Failure engineering and resilience testing\nDifficulty: advanced\nLanguage: Go\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nMetrics:\n- pass@k\n- time-to-green\n- cost-per-fix\n- regression rate\n", "metadata": {"language": "Go", "moe_experts": ["research", "evaluation"], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true}, "hash": "31c18fef10b18d3cb5838999ed873cd85ae5bd32"}
{"id": "train_039987", "created": "2026-01-01", "topic": "Multi-agent coordination and role separation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Multi-agent coordination and role separation\nDifficulty: expert\nLanguage: Bash\nContext: Offline air-gapped deployment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Bash", "moe_experts": ["security", "governance"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "2b52059ab5ffb3c0389192de1d8bd021ed28b81b"}
{"id": "train_039988", "created": "2026-01-01", "topic": "Model merging, distillation, and continual learning", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continual learning\nDifficulty: advanced\nLanguage: Java\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Java", "moe_experts": ["agent", "research"], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true}, "hash": "8237c0250266aceb46bfa9797a7f5a045a4a4972"}
{"id": "train_039989", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: Rust\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Rust", "moe_experts": ["performance", "governance"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "19655f0704e1eb746b9618b8eafff825b6179468"}
{"id": "train_039990", "created": "2026-01-01", "topic": "Governance, provenance, and audit logging", "task_type": "patch_diff", "difficulty": "advanced", "instruction": "Task: patch_diff\nTopic: Governance, provenance, and audit logging\nDifficulty: advanced\nLanguage: Rust\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Rust", "moe_experts": ["research", "security"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "cf44506cf60f5bef7d20f8c5321b6ca284b7553a"}
{"id": "train_039991", "created": "2026-01-01", "topic": "Self-grading and automated model critique", "task_type": "self_grade", "difficulty": "expert", "instruction": "Task: self_grade\nTopic: Self-grading and automated model critique\nDifficulty: expert\nLanguage: SQL\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\n{\n  \"score\": 6,\n  \"confidence\": 0.6,\n  \"notes\": \"Passed tests; minimal diff; no security flags.\"\n}", "metadata": {"language": "SQL", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "ec8db9a30a471ac6a3e45586e54d391bd5a8ed2d"}
{"id": "train_039992", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "patch_diff", "difficulty": "expert", "instruction": "Task: patch_diff\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: expert\nLanguage: Go\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nPatch:\n```diff\n- result = a / b\n+ if b == 0:\n+     raise ValueError('division by zero')\n+ result = a / b\n\n```", "metadata": {"language": "Go", "moe_experts": ["governance", "coding"], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true}, "hash": "d54f25a8b56feef01be01ecfc8d5417fe125433c"}
{"id": "train_039993", "created": "2026-01-01", "topic": "Secure code generation, red-teaming, and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation, red-teaming, and policy gates\nDifficulty: advanced\nLanguage: TypeScript\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["agent", "security"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "b36ffb7cc9cdbf3a3d3aed1f8a08e537b0287b37"}
{"id": "train_039994", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "failure_analysis", "difficulty": "advanced", "instruction": "Task: failure_analysis\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Bash\nContext: Regulated enterprise with audit requirements\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "Bash", "moe_experts": ["evaluation", "agent"], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true}, "hash": "67301215ccecd0b60cf19135ad9b37dc801a1091"}
{"id": "train_039995", "created": "2026-01-01", "topic": "Tool calling protocols and sandboxed execution", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling protocols and sandboxed execution\nDifficulty: expert\nLanguage: TypeScript\nContext: Large monorepo with flaky tests and strict CI\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "TypeScript", "moe_experts": ["governance", "data"], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true}, "hash": "20f4f08036fb94fe1f71e70709db468dc380bb85"}
{"id": "train_039996", "created": "2026-01-01", "topic": "Latency, cost, throughput optimization", "task_type": "failure_analysis", "difficulty": "expert", "instruction": "Task: failure_analysis\nTopic: Latency, cost, throughput optimization\nDifficulty: expert\nLanguage: JavaScript\nContext: Latency-critical production API\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nFailure: regression detected\nCause: missing edge case\nFix: add guard + test\n", "metadata": {"language": "JavaScript", "moe_experts": ["data", "performance"], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true}, "hash": "330d8be50cd12228596a54431c46372009fbfa5d"}
{"id": "train_039997", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["research", "evaluation"], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true}, "hash": "af4a641702e8aee3b28a5d8e2197a01e83a22d6e"}
{"id": "train_039998", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: expert\nLanguage: Python\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Python", "moe_experts": ["security", "data"], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true}, "hash": "f21db53ae89c2a83547d90d06f2f108fd47f287c"}
{"id": "train_039999", "created": "2026-01-01", "topic": "Extended context, memory, and repo graphs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context, memory, and repo graphs\nDifficulty: advanced\nLanguage: Go\nContext: Research benchmark validation environment\nProduce frontier-grade, production-valid output.", "input": "", "output": "Facts:\n- Modern AI coding success is measured by test pass rates, iteration cost, and reliability.\n- Agentic and multi-agent systems outperform single-pass generation.\n\nExpert guidance with risks, metrics, and governance controls.", "metadata": {"language": "Go", "moe_experts": ["security", "evaluation"], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true}, "hash": "98326ac977b60fa892536c0c7f919bf3e778ea77"}