{"id": "train_05000", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_05001", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tooling", "documentation"]}}
{"id": "train_05002", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tests_are_truth"]}}
{"id": "train_05003", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "governance", "security_gates"]}}
{"id": "train_05004", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "governance", "documentation"]}}
{"id": "train_05005", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "security_gates", "governance"]}}
{"id": "train_05006", "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\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "reproducibility"]}}
{"id": "train_05007", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "governance", "tooling"]}}
{"id": "train_05008", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "security_gates", "tests_are_truth"]}}
{"id": "train_05009", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "ci_integration", "reproducibility"]}}
{"id": "train_05010", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_05011", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_05012", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "documentation"]}}
{"id": "train_05013", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05014", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "security_gates", "governance"]}}
{"id": "train_05015", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_05016", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tooling", "ci_integration"]}}
{"id": "train_05017", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "documentation", "tests_are_truth"]}}
{"id": "train_05018", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "tests_are_truth", "security_gates"]}}
{"id": "train_05019", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "security_gates", "reproducibility"]}}
{"id": "train_05020", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "tests_are_truth"]}}
{"id": "train_05021", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tooling", "evaluation_metrics"]}}
{"id": "train_05022", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05023", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "security_gates", "reproducibility"]}}
{"id": "train_05024", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_05025", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tooling", "documentation"]}}
{"id": "train_05026", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "documentation", "ci_integration"]}}
{"id": "train_05027", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "tests_are_truth", "ci_integration"]}}
{"id": "train_05028", "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\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "documentation", "evaluation_metrics"]}}
{"id": "train_05029", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "documentation", "security_gates"]}}
{"id": "train_05030", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "evaluation_metrics", "documentation"]}}
{"id": "train_05031", "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\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "reproducibility", "ci_integration"]}}
{"id": "train_05032", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_05033", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "evaluation_metrics", "documentation"]}}
{"id": "train_05034", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tooling"]}}
{"id": "train_05035", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05036", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "security_gates", "tests_are_truth"]}}
{"id": "train_05037", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05038", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_05039", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "security_gates", "evaluation_metrics"]}}
{"id": "train_05040", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tests_are_truth", "ci_integration"]}}
{"id": "train_05041", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05042", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_05043", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "evaluation_metrics", "security_gates"]}}
{"id": "train_05044", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "documentation", "evaluation_metrics"]}}
{"id": "train_05045", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_05046", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05047", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_05048", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "reproducibility", "documentation"]}}
{"id": "train_05049", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_05050", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05051", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_05052", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tooling", "evaluation_metrics"]}}
{"id": "train_05053", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "evaluation_metrics", "reproducibility"]}}
{"id": "train_05054", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05055", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "reproducibility", "security_gates"]}}
{"id": "train_05056", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance"]}}
{"id": "train_05057", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_05058", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "reproducibility", "ci_integration"]}}
{"id": "train_05059", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_05060", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "reproducibility"]}}
{"id": "train_05061", "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\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05062", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates"]}}
{"id": "train_05063", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "governance", "tooling"]}}
{"id": "train_05064", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "governance", "reproducibility"]}}
{"id": "train_05065", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_05066", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tooling", "documentation"]}}
{"id": "train_05067", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "governance", "evaluation_metrics"]}}
{"id": "train_05068", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "security_gates", "governance"]}}
{"id": "train_05069", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "tests_are_truth"]}}
{"id": "train_05070", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "governance", "security_gates"]}}
{"id": "train_05071", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility"]}}
{"id": "train_05072", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_05073", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "documentation", "tests_are_truth"]}}
{"id": "train_05074", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_05075", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_05076", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "security_gates"]}}
{"id": "train_05077", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth"]}}
{"id": "train_05078", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "documentation", "tooling"]}}
{"id": "train_05079", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "security_gates", "evaluation_metrics"]}}
{"id": "train_05080", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "tooling"]}}
{"id": "train_05081", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "tests_are_truth"]}}
{"id": "train_05082", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "reproducibility", "governance"]}}
{"id": "train_05083", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "reproducibility", "documentation"]}}
{"id": "train_05084", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_05085", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_05086", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_05087", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "security_gates"]}}
{"id": "train_05088", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_05089", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "reproducibility"]}}
{"id": "train_05090", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tooling", "documentation"]}}
{"id": "train_05091", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tooling", "tests_are_truth"]}}
{"id": "train_05092", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tooling", "ci_integration"]}}
{"id": "train_05093", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth"]}}
{"id": "train_05094", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "reproducibility"]}}
{"id": "train_05095", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tests_are_truth", "governance"]}}
{"id": "train_05096", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "documentation", "evaluation_metrics"]}}
{"id": "train_05097", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "documentation", "tests_are_truth"]}}
{"id": "train_05098", "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\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_05099", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "ci_integration"]}}
{"id": "train_05100", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_05101", "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\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_05102", "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\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_05103", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_05104", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "documentation", "tests_are_truth"]}}
{"id": "train_05105", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "governance", "documentation"]}}
{"id": "train_05106", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "security_gates", "tooling"]}}
{"id": "train_05107", "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\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "evaluation_metrics", "tooling"]}}
{"id": "train_05108", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_05109", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "documentation", "evaluation_metrics"]}}
{"id": "train_05110", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "evaluation_metrics", "documentation"]}}
{"id": "train_05111", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "governance", "tests_are_truth"]}}
{"id": "train_05112", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "reproducibility", "ci_integration"]}}
{"id": "train_05113", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "security_gates", "governance"]}}
{"id": "train_05114", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "ci_integration", "governance"]}}
{"id": "train_05115", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "security_gates"]}}
{"id": "train_05116", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05117", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "ci_integration"]}}
{"id": "train_05118", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05119", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tests_are_truth", "tooling"]}}
{"id": "train_05120", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "ci_integration", "documentation"]}}
{"id": "train_05121", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_05122", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tooling", "security_gates"]}}
{"id": "train_05123", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "reproducibility", "tooling"]}}
{"id": "train_05124", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tooling", "ci_integration"]}}
{"id": "train_05125", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_05126", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "reproducibility", "tooling"]}}
{"id": "train_05127", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "documentation", "evaluation_metrics"]}}
{"id": "train_05128", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling"]}}
{"id": "train_05129", "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\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_05130", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "tests_are_truth", "governance"]}}
{"id": "train_05131", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "security_gates", "governance"]}}
{"id": "train_05132", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "governance", "repo_scale_reasoning"]}}
{"id": "train_05133", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_05134", "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\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_05135", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "repo_scale_reasoning", "documentation"]}}
{"id": "train_05136", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "governance", "documentation"]}}
{"id": "train_05137", "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\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "ci_integration"]}}
{"id": "train_05138", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "documentation", "evaluation_metrics"]}}
{"id": "train_05139", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "governance", "documentation"]}}
{"id": "train_05140", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "security_gates", "governance"]}}
{"id": "train_05141", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "tooling"]}}
{"id": "train_05142", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "security_gates", "documentation"]}}
{"id": "train_05143", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "governance", "documentation"]}}
{"id": "train_05144", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_05145", "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\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tooling", "repo_scale_reasoning"]}}
{"id": "train_05146", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "security_gates", "tooling"]}}
{"id": "train_05147", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "documentation", "repo_scale_reasoning"]}}
{"id": "train_05148", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "evaluation_metrics", "reproducibility"]}}
{"id": "train_05149", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "reproducibility", "governance"]}}
{"id": "train_05150", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "security_gates"]}}
{"id": "train_05151", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "documentation", "ci_integration"]}}
{"id": "train_05152", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "security_gates"]}}
{"id": "train_05153", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "ci_integration", "security_gates"]}}
{"id": "train_05154", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_05155", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05156", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "evaluation_metrics", "tooling"]}}
{"id": "train_05157", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility"]}}
{"id": "train_05158", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "documentation", "ci_integration"]}}
{"id": "train_05159", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_05160", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "ci_integration", "evaluation_metrics"]}}
{"id": "train_05161", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "governance", "security_gates"]}}
{"id": "train_05162", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_05163", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_05164", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "tests_are_truth", "security_gates"]}}
{"id": "train_05165", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "ci_integration", "governance"]}}
{"id": "train_05166", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05167", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "security_gates", "tests_are_truth"]}}
{"id": "train_05168", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning"]}}
{"id": "train_05169", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "reproducibility", "tooling"]}}
{"id": "train_05170", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "documentation", "security_gates"]}}
{"id": "train_05171", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "governance", "documentation"]}}
{"id": "train_05172", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "security_gates", "documentation"]}}
{"id": "train_05173", "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\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "reproducibility", "documentation"]}}
{"id": "train_05174", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_05175", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_05176", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05177", "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\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "security_gates", "governance"]}}
{"id": "train_05178", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_05179", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "documentation", "evaluation_metrics"]}}
{"id": "train_05180", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "tests_are_truth"]}}
{"id": "train_05181", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_05182", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "governance", "ci_integration"]}}
{"id": "train_05183", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tooling", "repo_scale_reasoning"]}}
{"id": "train_05184", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tooling", "tests_are_truth"]}}
{"id": "train_05185", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "documentation", "evaluation_metrics"]}}
{"id": "train_05186", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "documentation", "evaluation_metrics"]}}
{"id": "train_05187", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_05188", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "repo_scale_reasoning", "governance"]}}
{"id": "train_05189", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tooling", "evaluation_metrics"]}}
{"id": "train_05190", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_05191", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "governance"]}}
{"id": "train_05192", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05193", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "tooling", "governance"]}}
{"id": "train_05194", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "security_gates", "tests_are_truth"]}}
{"id": "train_05195", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_05196", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_05197", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tooling", "documentation"]}}
{"id": "train_05198", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "governance", "reproducibility"]}}
{"id": "train_05199", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_05200", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "ci_integration", "reproducibility"]}}
{"id": "train_05201", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "reproducibility", "governance"]}}
{"id": "train_05202", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "ci_integration", "documentation"]}}
{"id": "train_05203", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "documentation", "repo_scale_reasoning"]}}
{"id": "train_05204", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tooling", "ci_integration"]}}
{"id": "train_05205", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tests_are_truth", "tooling"]}}
{"id": "train_05206", "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\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_05207", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "ci_integration"]}}
{"id": "train_05208", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05209", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tooling", "ci_integration"]}}
{"id": "train_05210", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "security_gates", "evaluation_metrics"]}}
{"id": "train_05211", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_05212", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tooling", "repo_scale_reasoning"]}}
{"id": "train_05213", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "ci_integration", "tests_are_truth"]}}
{"id": "train_05214", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "governance"]}}
{"id": "train_05215", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05216", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "tooling"]}}
{"id": "train_05217", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "security_gates", "tooling"]}}
{"id": "train_05218", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_05219", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_05220", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "tooling"]}}
{"id": "train_05221", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "ci_integration", "tooling"]}}
{"id": "train_05222", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tooling", "ci_integration"]}}
{"id": "train_05223", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_05224", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_05225", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_05226", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_05227", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "security_gates", "documentation"]}}
{"id": "train_05228", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "governance", "tooling"]}}
{"id": "train_05229", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tooling", "reproducibility"]}}
{"id": "train_05230", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "ci_integration", "tooling"]}}
{"id": "train_05231", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "governance"]}}
{"id": "train_05232", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "governance"]}}
{"id": "train_05233", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "security_gates", "governance"]}}
{"id": "train_05234", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_05235", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "documentation", "evaluation_metrics"]}}
{"id": "train_05236", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tooling", "reproducibility"]}}
{"id": "train_05237", "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\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_05238", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tooling"]}}
{"id": "train_05239", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tooling"]}}
{"id": "train_05240", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "documentation", "tests_are_truth"]}}
{"id": "train_05241", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "evaluation_metrics"]}}
{"id": "train_05242", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "governance", "ci_integration"]}}
{"id": "train_05243", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_05244", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_05245", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "tooling"]}}
{"id": "train_05246", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_05247", "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\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05248", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "reproducibility", "governance"]}}
{"id": "train_05249", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_05250", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tooling", "ci_integration"]}}
{"id": "train_05251", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "security_gates", "tests_are_truth"]}}
{"id": "train_05252", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_05253", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_05254", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_05255", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tooling", "repo_scale_reasoning"]}}
{"id": "train_05256", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics"]}}
{"id": "train_05257", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_05258", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_05259", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_05260", "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\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "security_gates", "ci_integration"]}}
{"id": "train_05261", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tooling", "security_gates"]}}
{"id": "train_05262", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "repo_scale_reasoning"]}}
{"id": "train_05263", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "evaluation_metrics"]}}
{"id": "train_05264", "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\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tests_are_truth", "documentation"]}}
{"id": "train_05265", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "documentation", "evaluation_metrics"]}}
{"id": "train_05266", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "reproducibility"]}}
{"id": "train_05267", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "tests_are_truth"]}}
{"id": "train_05268", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "documentation", "security_gates"]}}
{"id": "train_05269", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tooling", "tests_are_truth"]}}
{"id": "train_05270", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_05271", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "reproducibility", "documentation"]}}
{"id": "train_05272", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "reproducibility", "governance"]}}
{"id": "train_05273", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_05274", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "security_gates"]}}
{"id": "train_05275", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "security_gates"]}}
{"id": "train_05276", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_05277", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_05278", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_05279", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "governance", "reproducibility"]}}
{"id": "train_05280", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "tooling", "repo_scale_reasoning"]}}
{"id": "train_05281", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_05282", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "governance"]}}
{"id": "train_05283", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tests_are_truth", "security_gates"]}}
{"id": "train_05284", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth"]}}
{"id": "train_05285", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_05286", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "documentation"]}}
{"id": "train_05287", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "governance", "tooling"]}}
{"id": "train_05288", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "ci_integration"]}}
{"id": "train_05289", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_05290", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "reproducibility", "security_gates"]}}
{"id": "train_05291", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_05292", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05293", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "reproducibility"]}}
{"id": "train_05294", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_05295", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_05296", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "security_gates", "ci_integration"]}}
{"id": "train_05297", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_05298", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_05299", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_05300", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "documentation"]}}
{"id": "train_05301", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility"]}}
{"id": "train_05302", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_05303", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_05304", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_05305", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tooling", "repo_scale_reasoning"]}}
{"id": "train_05306", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_05307", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tooling"]}}
{"id": "train_05308", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "evaluation_metrics", "reproducibility"]}}
{"id": "train_05309", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "tooling"]}}
{"id": "train_05310", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05311", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "security_gates", "documentation"]}}
{"id": "train_05312", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05313", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "repo_scale_reasoning", "governance"]}}
{"id": "train_05314", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tooling"]}}
{"id": "train_05315", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "security_gates", "tooling"]}}
{"id": "train_05316", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "governance", "reproducibility"]}}
{"id": "train_05317", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_05318", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "evaluation_metrics", "security_gates"]}}
{"id": "train_05319", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "governance", "ci_integration"]}}
{"id": "train_05320", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "reproducibility", "governance"]}}
{"id": "train_05321", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tests_are_truth", "documentation"]}}
{"id": "train_05322", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_05323", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tooling", "evaluation_metrics"]}}
{"id": "train_05324", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "governance", "tests_are_truth"]}}
{"id": "train_05325", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "documentation", "tests_are_truth"]}}
{"id": "train_05326", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05327", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tests_are_truth", "documentation"]}}
{"id": "train_05328", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "documentation", "ci_integration"]}}
{"id": "train_05329", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_05330", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "governance", "reproducibility"]}}
{"id": "train_05331", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_05332", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "governance", "security_gates"]}}
{"id": "train_05333", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_05334", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tooling", "reproducibility"]}}
{"id": "train_05335", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration"]}}
{"id": "train_05336", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "security_gates", "ci_integration"]}}
{"id": "train_05337", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "tooling"]}}
{"id": "train_05338", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "security_gates", "governance"]}}
{"id": "train_05339", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "documentation", "governance"]}}
{"id": "train_05340", "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\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "tooling"]}}
{"id": "train_05341", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "documentation", "tooling"]}}
{"id": "train_05342", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_05343", "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\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling"]}}
{"id": "train_05344", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "evaluation_metrics"]}}
{"id": "train_05345", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_05346", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_05347", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_05348", "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\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05349", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05350", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "evaluation_metrics", "tooling"]}}
{"id": "train_05351", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "governance", "tooling"]}}
{"id": "train_05352", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tooling", "evaluation_metrics"]}}
{"id": "train_05353", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_05354", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "governance", "documentation"]}}
{"id": "train_05355", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "evaluation_metrics", "reproducibility"]}}
{"id": "train_05356", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics"]}}
{"id": "train_05357", "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\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_05358", "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\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation"]}}
{"id": "train_05359", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "governance"]}}
{"id": "train_05360", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05361", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_05362", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "reproducibility", "security_gates"]}}
{"id": "train_05363", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "security_gates"]}}
{"id": "train_05364", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "evaluation_metrics", "documentation"]}}
{"id": "train_05365", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth"]}}
{"id": "train_05366", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "documentation", "tests_are_truth"]}}
{"id": "train_05367", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth"]}}
{"id": "train_05368", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "documentation", "repo_scale_reasoning"]}}
{"id": "train_05369", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05370", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tooling", "repo_scale_reasoning"]}}
{"id": "train_05371", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_05372", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "governance", "tests_are_truth"]}}
{"id": "train_05373", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling"]}}
{"id": "train_05374", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "governance", "ci_integration"]}}
{"id": "train_05375", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "reproducibility", "security_gates"]}}
{"id": "train_05376", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "governance", "reproducibility"]}}
{"id": "train_05377", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_05378", "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\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "security_gates", "evaluation_metrics"]}}
{"id": "train_05379", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "governance"]}}
{"id": "train_05380", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_05381", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "evaluation_metrics", "tooling"]}}
{"id": "train_05382", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_05383", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_05384", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "repo_scale_reasoning", "governance"]}}
{"id": "train_05385", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "ci_integration", "evaluation_metrics"]}}
{"id": "train_05386", "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\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05387", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "governance", "tests_are_truth"]}}
{"id": "train_05388", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tests_are_truth", "reproducibility"]}}
{"id": "train_05389", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation"]}}
{"id": "train_05390", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tooling", "security_gates"]}}
{"id": "train_05391", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_05392", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_05393", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_05394", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_05395", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_05396", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "documentation", "evaluation_metrics"]}}
{"id": "train_05397", "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\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "tooling", "repo_scale_reasoning"]}}
{"id": "train_05398", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05399", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_05400", "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\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "governance", "tooling"]}}
{"id": "train_05401", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_05402", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tooling", "security_gates"]}}
{"id": "train_05403", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tooling"]}}
{"id": "train_05404", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling"]}}
{"id": "train_05405", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "repo_scale_reasoning", "documentation"]}}
{"id": "train_05406", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_05407", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05408", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_05409", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_05410", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_05411", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05412", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_05413", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tooling", "reproducibility"]}}
{"id": "train_05414", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05415", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_05416", "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\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tooling", "security_gates"]}}
{"id": "train_05417", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "governance", "evaluation_metrics"]}}
{"id": "train_05418", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_05419", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "reproducibility", "tooling"]}}
{"id": "train_05420", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "governance", "repo_scale_reasoning"]}}
{"id": "train_05421", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_05422", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "evaluation_metrics"]}}
{"id": "train_05423", "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\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "tooling"]}}
{"id": "train_05424", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "evaluation_metrics", "ci_integration"]}}
{"id": "train_05425", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tooling", "governance"]}}
{"id": "train_05426", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "documentation", "ci_integration"]}}
{"id": "train_05427", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "security_gates", "ci_integration"]}}
{"id": "train_05428", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05429", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05430", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics"]}}
{"id": "train_05431", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tests_are_truth", "documentation"]}}
{"id": "train_05432", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_05433", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "governance"]}}
{"id": "train_05434", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tooling", "security_gates"]}}
{"id": "train_05435", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "ci_integration", "tests_are_truth"]}}
{"id": "train_05436", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tooling", "governance"]}}
{"id": "train_05437", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_05438", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "governance", "reproducibility"]}}
{"id": "train_05439", "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\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance"]}}
{"id": "train_05440", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "evaluation_metrics", "tooling"]}}
{"id": "train_05441", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05442", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_05443", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05444", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_05445", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_05446", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_05447", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "tooling"]}}
{"id": "train_05448", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "documentation", "tooling"]}}
{"id": "train_05449", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "tooling"]}}
{"id": "train_05450", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_05451", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_05452", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "documentation", "governance"]}}
{"id": "train_05453", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "governance", "ci_integration"]}}
{"id": "train_05454", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "tests_are_truth"]}}
{"id": "train_05455", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_05456", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "governance", "tests_are_truth"]}}
{"id": "train_05457", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "ci_integration", "evaluation_metrics"]}}
{"id": "train_05458", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "evaluation_metrics", "governance"]}}
{"id": "train_05459", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_05460", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "documentation", "ci_integration"]}}
{"id": "train_05461", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "governance", "tests_are_truth"]}}
{"id": "train_05462", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning"]}}
{"id": "train_05463", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "security_gates", "tooling"]}}
{"id": "train_05464", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "ci_integration", "governance"]}}
{"id": "train_05465", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "ci_integration", "security_gates"]}}
{"id": "train_05466", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tests_are_truth", "governance"]}}
{"id": "train_05467", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "evaluation_metrics", "tooling"]}}
{"id": "train_05468", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_05469", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05470", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "documentation", "tests_are_truth"]}}
{"id": "train_05471", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "ci_integration", "tooling"]}}
{"id": "train_05472", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "repo_scale_reasoning", "governance"]}}
{"id": "train_05473", "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\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_05474", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tooling", "governance"]}}
{"id": "train_05475", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "ci_integration", "evaluation_metrics"]}}
{"id": "train_05476", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tooling"]}}
{"id": "train_05477", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "security_gates", "tooling"]}}
{"id": "train_05478", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_05479", "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\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tooling", "repo_scale_reasoning"]}}
{"id": "train_05480", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "governance", "reproducibility"]}}
{"id": "train_05481", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "reproducibility"]}}
{"id": "train_05482", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_05483", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "documentation", "security_gates"]}}
{"id": "train_05484", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "documentation"]}}
{"id": "train_05485", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance"]}}
{"id": "train_05486", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates"]}}
{"id": "train_05487", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "ci_integration", "evaluation_metrics"]}}
{"id": "train_05488", "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\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "security_gates", "governance"]}}
{"id": "train_05489", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "governance"]}}
{"id": "train_05490", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "ci_integration", "tooling"]}}
{"id": "train_05491", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_05492", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05493", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_05494", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "ci_integration", "tooling"]}}
{"id": "train_05495", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_05496", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "security_gates", "evaluation_metrics"]}}
{"id": "train_05497", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "documentation", "tooling"]}}
{"id": "train_05498", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "reproducibility"]}}
{"id": "train_05499", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "ci_integration", "evaluation_metrics"]}}
{"id": "train_05500", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "governance"]}}
{"id": "train_05501", "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\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "ci_integration", "tests_are_truth"]}}
{"id": "train_05502", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "governance", "documentation"]}}
{"id": "train_05503", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "security_gates", "governance"]}}
{"id": "train_05504", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "documentation", "tests_are_truth"]}}
{"id": "train_05505", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05506", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05507", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "governance", "documentation"]}}
{"id": "train_05508", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05509", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_05510", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05511", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "evaluation_metrics", "governance"]}}
{"id": "train_05512", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "ci_integration"]}}
{"id": "train_05513", "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\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "evaluation_metrics", "ci_integration"]}}
{"id": "train_05514", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tooling", "governance"]}}
{"id": "train_05515", "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\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "documentation", "tests_are_truth"]}}
{"id": "train_05516", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_05517", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_05518", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics"]}}
{"id": "train_05519", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tooling", "ci_integration"]}}
{"id": "train_05520", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "tooling", "repo_scale_reasoning"]}}
{"id": "train_05521", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "reproducibility"]}}
{"id": "train_05522", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "repo_scale_reasoning", "tooling"]}}
{"id": "train_05523", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tooling", "governance"]}}
{"id": "train_05524", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_05525", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_05526", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05527", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05528", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "tooling"]}}
{"id": "train_05529", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates"]}}
{"id": "train_05530", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "security_gates", "evaluation_metrics"]}}
{"id": "train_05531", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth"]}}
{"id": "train_05532", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tooling", "reproducibility"]}}
{"id": "train_05533", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "security_gates", "tooling"]}}
{"id": "train_05534", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tooling", "security_gates"]}}
{"id": "train_05535", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "documentation", "ci_integration"]}}
{"id": "train_05536", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "ci_integration", "evaluation_metrics"]}}
{"id": "train_05537", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tests_are_truth", "reproducibility"]}}
{"id": "train_05538", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "documentation", "governance"]}}
{"id": "train_05539", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05540", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "reproducibility", "governance"]}}
{"id": "train_05541", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "ci_integration"]}}
{"id": "train_05542", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_05543", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "reproducibility", "documentation"]}}
{"id": "train_05544", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "governance"]}}
{"id": "train_05545", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "governance", "tests_are_truth"]}}
{"id": "train_05546", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "ci_integration", "documentation"]}}
{"id": "train_05547", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "ci_integration", "reproducibility"]}}
{"id": "train_05548", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "ci_integration", "reproducibility"]}}
{"id": "train_05549", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_05550", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tests_are_truth", "security_gates"]}}
{"id": "train_05551", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05552", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_05553", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_05554", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "ci_integration", "governance"]}}
{"id": "train_05555", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "security_gates", "documentation"]}}
{"id": "train_05556", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "reproducibility", "security_gates"]}}
{"id": "train_05557", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_05558", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning"]}}
{"id": "train_05559", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "governance", "tooling"]}}
{"id": "train_05560", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "documentation", "evaluation_metrics"]}}
{"id": "train_05561", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics"]}}
{"id": "train_05562", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "security_gates"]}}
{"id": "train_05563", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "reproducibility", "tests_are_truth"]}}
{"id": "train_05564", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_05565", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tooling", "reproducibility"]}}
{"id": "train_05566", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_05567", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "documentation"]}}
{"id": "train_05568", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "security_gates", "reproducibility"]}}
{"id": "train_05569", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_05570", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "repo_scale_reasoning", "governance"]}}
{"id": "train_05571", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "security_gates", "documentation"]}}
{"id": "train_05572", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "evaluation_metrics", "documentation"]}}
{"id": "train_05573", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_05574", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "security_gates", "reproducibility"]}}
{"id": "train_05575", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "documentation", "tooling"]}}
{"id": "train_05576", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "documentation", "ci_integration"]}}
{"id": "train_05577", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth"]}}
{"id": "train_05578", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation"]}}
{"id": "train_05579", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tests_are_truth", "security_gates"]}}
{"id": "train_05580", "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\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_05581", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "tooling", "documentation"]}}
{"id": "train_05582", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_05583", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "governance", "ci_integration"]}}
{"id": "train_05584", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_05585", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "security_gates", "tooling"]}}
{"id": "train_05586", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "governance", "tests_are_truth"]}}
{"id": "train_05587", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_05588", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tooling", "evaluation_metrics"]}}
{"id": "train_05589", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "ci_integration"]}}
{"id": "train_05590", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tooling", "ci_integration"]}}
{"id": "train_05591", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "documentation", "repo_scale_reasoning"]}}
{"id": "train_05592", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tooling", "tests_are_truth"]}}
{"id": "train_05593", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_05594", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "evaluation_metrics", "tooling"]}}
{"id": "train_05595", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05596", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_05597", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_05598", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05599", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_05600", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_05601", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_05602", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "governance", "evaluation_metrics"]}}
{"id": "train_05603", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "reproducibility", "ci_integration"]}}
{"id": "train_05604", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "documentation", "repo_scale_reasoning"]}}
{"id": "train_05605", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "ci_integration", "reproducibility"]}}
{"id": "train_05606", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "ci_integration", "governance"]}}
{"id": "train_05607", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "tooling"]}}
{"id": "train_05608", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_05609", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "documentation"]}}
{"id": "train_05610", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_05611", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "reproducibility", "governance"]}}
{"id": "train_05612", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "security_gates"]}}
{"id": "train_05613", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_05614", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "ci_integration", "evaluation_metrics"]}}
{"id": "train_05615", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_05616", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "security_gates", "ci_integration"]}}
{"id": "train_05617", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "tooling", "ci_integration"]}}
{"id": "train_05618", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "governance", "evaluation_metrics"]}}
{"id": "train_05619", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tooling", "governance"]}}
{"id": "train_05620", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_05621", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_05622", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "tests_are_truth", "governance"]}}
{"id": "train_05623", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_05624", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "reproducibility", "governance"]}}
{"id": "train_05625", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_05626", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05627", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_05628", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tooling", "tests_are_truth"]}}
{"id": "train_05629", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "documentation", "reproducibility"]}}
{"id": "train_05630", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tests_are_truth", "security_gates"]}}
{"id": "train_05631", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "tests_are_truth"]}}
{"id": "train_05632", "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\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tests_are_truth", "reproducibility"]}}
{"id": "train_05633", "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\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "documentation", "evaluation_metrics"]}}
{"id": "train_05634", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tooling", "tests_are_truth"]}}
{"id": "train_05635", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "security_gates", "tests_are_truth"]}}
{"id": "train_05636", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "governance", "tests_are_truth"]}}
{"id": "train_05637", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "documentation", "ci_integration"]}}
{"id": "train_05638", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tooling", "security_gates"]}}
{"id": "train_05639", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation"]}}
{"id": "train_05640", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "evaluation_metrics", "security_gates"]}}
{"id": "train_05641", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_05642", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates"]}}
{"id": "train_05643", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "governance", "tests_are_truth"]}}
{"id": "train_05644", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth"]}}
{"id": "train_05645", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05646", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05647", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05648", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance"]}}
{"id": "train_05649", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_05650", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning"]}}
{"id": "train_05651", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_05652", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_05653", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "ci_integration"]}}
{"id": "train_05654", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05655", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_05656", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "security_gates", "evaluation_metrics"]}}
{"id": "train_05657", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "security_gates", "reproducibility"]}}
{"id": "train_05658", "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\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "documentation", "security_gates"]}}
{"id": "train_05659", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_05660", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "evaluation_metrics", "reproducibility"]}}
{"id": "train_05661", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_05662", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05663", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "ci_integration", "governance"]}}
{"id": "train_05664", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "governance"]}}
{"id": "train_05665", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "security_gates"]}}
{"id": "train_05666", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_05667", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05668", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_05669", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tests_are_truth", "ci_integration"]}}
{"id": "train_05670", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "tests_are_truth"]}}
{"id": "train_05671", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_05672", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "ci_integration"]}}
{"id": "train_05673", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates"]}}
{"id": "train_05674", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_05675", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "evaluation_metrics", "tooling"]}}
{"id": "train_05676", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates"]}}
{"id": "train_05677", "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\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "security_gates"]}}
{"id": "train_05678", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05679", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05680", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "governance"]}}
{"id": "train_05681", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "reproducibility", "ci_integration"]}}
{"id": "train_05682", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "security_gates", "tooling"]}}
{"id": "train_05683", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05684", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tests_are_truth", "governance"]}}
{"id": "train_05685", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "documentation", "security_gates"]}}
{"id": "train_05686", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_05687", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tooling", "reproducibility"]}}
{"id": "train_05688", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "evaluation_metrics"]}}
{"id": "train_05689", "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\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "documentation", "security_gates"]}}
{"id": "train_05690", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "ci_integration", "documentation"]}}
{"id": "train_05691", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "security_gates", "tooling"]}}
{"id": "train_05692", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "governance", "security_gates"]}}
{"id": "train_05693", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tooling", "tests_are_truth"]}}
{"id": "train_05694", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "ci_integration", "security_gates"]}}
{"id": "train_05695", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "documentation", "reproducibility"]}}
{"id": "train_05696", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_05697", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "security_gates", "reproducibility"]}}
{"id": "train_05698", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "documentation", "reproducibility"]}}
{"id": "train_05699", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics"]}}
{"id": "train_05700", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_05701", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "governance"]}}
{"id": "train_05702", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "documentation"]}}
{"id": "train_05703", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05704", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tests_are_truth", "security_gates"]}}
{"id": "train_05705", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_05706", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "governance", "ci_integration"]}}
{"id": "train_05707", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tooling", "tests_are_truth"]}}
{"id": "train_05708", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "governance", "reproducibility"]}}
{"id": "train_05709", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "governance"]}}
{"id": "train_05710", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tooling", "documentation"]}}
{"id": "train_05711", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05712", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "tests_are_truth"]}}
{"id": "train_05713", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_05714", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "security_gates"]}}
{"id": "train_05715", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05716", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "tooling"]}}
{"id": "train_05717", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_05718", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_05719", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_05720", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth"]}}
{"id": "train_05721", "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\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "ci_integration", "evaluation_metrics"]}}
{"id": "train_05722", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_05723", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "tests_are_truth", "ci_integration"]}}
{"id": "train_05724", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "tooling"]}}
{"id": "train_05725", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_05726", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "documentation"]}}
{"id": "train_05727", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_05728", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_05729", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05730", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "documentation"]}}
{"id": "train_05731", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_05732", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tooling"]}}
{"id": "train_05733", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "reproducibility", "governance"]}}
{"id": "train_05734", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "governance"]}}
{"id": "train_05735", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "documentation"]}}
{"id": "train_05736", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_05737", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_05738", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_05739", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "documentation", "tooling"]}}
{"id": "train_05740", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "documentation", "security_gates"]}}
{"id": "train_05741", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_05742", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "documentation", "reproducibility"]}}
{"id": "train_05743", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_05744", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "tests_are_truth", "security_gates"]}}
{"id": "train_05745", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_05746", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_05747", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "reproducibility", "documentation"]}}
{"id": "train_05748", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "governance", "repo_scale_reasoning"]}}
{"id": "train_05749", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_05750", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_05751", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tooling"]}}
{"id": "train_05752", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "ci_integration", "documentation"]}}
{"id": "train_05753", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_05754", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "reproducibility", "documentation"]}}
{"id": "train_05755", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "governance"]}}
{"id": "train_05756", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "ci_integration"]}}
{"id": "train_05757", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05758", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_05759", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05760", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "reproducibility", "governance"]}}
{"id": "train_05761", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "evaluation_metrics", "ci_integration"]}}
{"id": "train_05762", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_05763", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "ci_integration", "security_gates"]}}
{"id": "train_05764", "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\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "reproducibility"]}}
{"id": "train_05765", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "documentation", "repo_scale_reasoning"]}}
{"id": "train_05766", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_05767", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_05768", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "governance", "tooling"]}}
{"id": "train_05769", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "documentation", "reproducibility"]}}
{"id": "train_05770", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "documentation", "reproducibility"]}}
{"id": "train_05771", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "reproducibility", "governance"]}}
{"id": "train_05772", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "governance"]}}
{"id": "train_05773", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_05774", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05775", "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\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "evaluation_metrics", "documentation"]}}
{"id": "train_05776", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tooling", "reproducibility"]}}
{"id": "train_05777", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tests_are_truth", "governance"]}}
{"id": "train_05778", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "reproducibility", "security_gates"]}}
{"id": "train_05779", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth"]}}
{"id": "train_05780", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "governance"]}}
{"id": "train_05781", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "ci_integration", "reproducibility"]}}
{"id": "train_05782", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "governance", "documentation"]}}
{"id": "train_05783", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tests_are_truth", "reproducibility"]}}
{"id": "train_05784", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "documentation", "reproducibility"]}}
{"id": "train_05785", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_05786", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "governance", "ci_integration"]}}
{"id": "train_05787", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05788", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_05789", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tests_are_truth", "governance"]}}
{"id": "train_05790", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "governance", "reproducibility"]}}
{"id": "train_05791", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "governance", "ci_integration"]}}
{"id": "train_05792", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "documentation", "repo_scale_reasoning"]}}
{"id": "train_05793", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tests_are_truth", "tooling"]}}
{"id": "train_05794", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_05795", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tooling", "governance"]}}
{"id": "train_05796", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "security_gates"]}}
{"id": "train_05797", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning"]}}
{"id": "train_05798", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_05799", "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\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tooling", "reproducibility"]}}
{"id": "train_05800", "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\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "governance", "documentation"]}}
{"id": "train_05801", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tooling"]}}
{"id": "train_05802", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tooling"]}}
{"id": "train_05803", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tooling", "governance"]}}
{"id": "train_05804", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tests_are_truth", "reproducibility"]}}
{"id": "train_05805", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "ci_integration", "tests_are_truth"]}}
{"id": "train_05806", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_05807", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "tooling", "reproducibility"]}}
{"id": "train_05808", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics"]}}
{"id": "train_05809", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "ci_integration", "governance"]}}
{"id": "train_05810", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "tests_are_truth"]}}
{"id": "train_05811", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05812", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "reproducibility", "governance"]}}
{"id": "train_05813", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_05814", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_05815", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "evaluation_metrics"]}}
{"id": "train_05816", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "ci_integration", "documentation"]}}
{"id": "train_05817", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "reproducibility", "governance"]}}
{"id": "train_05818", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05819", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tests_are_truth", "security_gates"]}}
{"id": "train_05820", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_05821", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "reproducibility", "tests_are_truth"]}}
{"id": "train_05822", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tooling"]}}
{"id": "train_05823", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_05824", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_05825", "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\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tests_are_truth", "tooling"]}}
{"id": "train_05826", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tests_are_truth", "governance"]}}
{"id": "train_05827", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tooling", "reproducibility"]}}
{"id": "train_05828", "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\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "tests_are_truth", "tooling"]}}
{"id": "train_05829", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_05830", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tooling", "evaluation_metrics"]}}
{"id": "train_05831", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "tooling", "ci_integration"]}}
{"id": "train_05832", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "governance", "documentation"]}}
{"id": "train_05833", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05834", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05835", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_05836", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "reproducibility", "security_gates"]}}
{"id": "train_05837", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics"]}}
{"id": "train_05838", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "documentation", "tests_are_truth"]}}
{"id": "train_05839", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_05840", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "governance", "reproducibility"]}}
{"id": "train_05841", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tests_are_truth", "documentation"]}}
{"id": "train_05842", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_05843", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "governance"]}}
{"id": "train_05844", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05845", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_05846", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_05847", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "governance", "reproducibility"]}}
{"id": "train_05848", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_05849", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_05850", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tests_are_truth", "security_gates"]}}
{"id": "train_05851", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "documentation", "security_gates"]}}
{"id": "train_05852", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics"]}}
{"id": "train_05853", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "tests_are_truth", "security_gates"]}}
{"id": "train_05854", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "ci_integration", "tooling"]}}
{"id": "train_05855", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_05856", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling"]}}
{"id": "train_05857", "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\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "ci_integration", "security_gates"]}}
{"id": "train_05858", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "tests_are_truth"]}}
{"id": "train_05859", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05860", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05861", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_05862", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05863", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "repo_scale_reasoning", "tooling"]}}
{"id": "train_05864", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_05865", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_05866", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tests_are_truth", "documentation"]}}
{"id": "train_05867", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05868", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tests_are_truth"]}}
{"id": "train_05869", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05870", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05871", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_05872", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "evaluation_metrics"]}}
{"id": "train_05873", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "ci_integration", "documentation"]}}
{"id": "train_05874", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "evaluation_metrics", "reproducibility"]}}
{"id": "train_05875", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "security_gates", "evaluation_metrics"]}}
{"id": "train_05876", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_05877", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05878", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_05879", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "security_gates", "evaluation_metrics"]}}
{"id": "train_05880", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration"]}}
{"id": "train_05881", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_05882", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_05883", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "security_gates", "evaluation_metrics"]}}
{"id": "train_05884", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "evaluation_metrics", "ci_integration"]}}
{"id": "train_05885", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "repo_scale_reasoning", "tooling"]}}
{"id": "train_05886", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "tooling"]}}
{"id": "train_05887", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "reproducibility"]}}
{"id": "train_05888", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05889", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_05890", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "reproducibility", "governance"]}}
{"id": "train_05891", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tooling", "evaluation_metrics"]}}
{"id": "train_05892", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tests_are_truth", "governance"]}}
{"id": "train_05893", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tests_are_truth", "security_gates"]}}
{"id": "train_05894", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "ci_integration"]}}
{"id": "train_05895", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_05896", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "ci_integration", "evaluation_metrics"]}}
{"id": "train_05897", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "repo_scale_reasoning", "governance"]}}
{"id": "train_05898", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "documentation"]}}
{"id": "train_05899", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "tests_are_truth", "ci_integration"]}}
{"id": "train_05900", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "governance"]}}
{"id": "train_05901", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_05902", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "documentation", "tooling"]}}
{"id": "train_05903", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_05904", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "ci_integration"]}}
{"id": "train_05905", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_05906", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "evaluation_metrics", "tooling"]}}
{"id": "train_05907", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "governance", "evaluation_metrics"]}}
{"id": "train_05908", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "evaluation_metrics", "governance"]}}
{"id": "train_05909", "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\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_05910", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_05911", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "security_gates", "tests_are_truth"]}}
{"id": "train_05912", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "evaluation_metrics", "ci_integration"]}}
{"id": "train_05913", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "documentation"]}}
{"id": "train_05914", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_05915", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "documentation", "ci_integration"]}}
{"id": "train_05916", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "documentation"]}}
{"id": "train_05917", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tooling", "documentation"]}}
{"id": "train_05918", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "reproducibility", "documentation"]}}
{"id": "train_05919", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "ci_integration", "governance"]}}
{"id": "train_05920", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05921", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning"]}}
{"id": "train_05922", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "reproducibility", "governance"]}}
{"id": "train_05923", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_05924", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tooling"]}}
{"id": "train_05925", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tooling", "reproducibility"]}}
{"id": "train_05926", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_05927", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "reproducibility"]}}
{"id": "train_05928", "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\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_05929", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "tests_are_truth", "reproducibility"]}}
{"id": "train_05930", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "documentation", "ci_integration"]}}
{"id": "train_05931", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tests_are_truth", "security_gates"]}}
{"id": "train_05932", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "governance"]}}
{"id": "train_05933", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance"]}}
{"id": "train_05934", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_05935", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "reproducibility", "security_gates"]}}
{"id": "train_05936", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "documentation"]}}
{"id": "train_05937", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_05938", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_05939", "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\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tooling", "documentation"]}}
{"id": "train_05940", "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\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tests_are_truth", "security_gates"]}}
{"id": "train_05941", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "ci_integration", "governance"]}}
{"id": "train_05942", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_05943", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "security_gates", "tooling"]}}
{"id": "train_05944", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "security_gates", "reproducibility"]}}
{"id": "train_05945", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05946", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "reproducibility", "ci_integration"]}}
{"id": "train_05947", "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\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_05948", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tooling", "ci_integration"]}}
{"id": "train_05949", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration"]}}
{"id": "train_05950", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tooling", "reproducibility"]}}
{"id": "train_05951", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "tooling", "ci_integration"]}}
{"id": "train_05952", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "security_gates", "tests_are_truth"]}}
{"id": "train_05953", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tests_are_truth", "security_gates"]}}
{"id": "train_05954", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "documentation", "tests_are_truth"]}}
{"id": "train_05955", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_05956", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "ci_integration"]}}
{"id": "train_05957", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "security_gates", "governance"]}}
{"id": "train_05958", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_05959", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "governance", "tests_are_truth"]}}
{"id": "train_05960", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "reproducibility"]}}
{"id": "train_05961", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "evaluation_metrics", "reproducibility"]}}
{"id": "train_05962", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_05963", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05964", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "evaluation_metrics", "governance"]}}
{"id": "train_05965", "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\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "ci_integration", "tooling"]}}
{"id": "train_05966", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "security_gates", "governance"]}}
{"id": "train_05967", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_05968", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_05969", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "tests_are_truth"]}}
{"id": "train_05970", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "ci_integration"]}}
{"id": "train_05971", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tooling", "evaluation_metrics"]}}
{"id": "train_05972", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tooling", "repo_scale_reasoning"]}}
{"id": "train_05973", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning"]}}
{"id": "train_05974", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "reproducibility", "ci_integration"]}}
{"id": "train_05975", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "governance", "tests_are_truth"]}}
{"id": "train_05976", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "security_gates", "tooling"]}}
{"id": "train_05977", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics"]}}
{"id": "train_05978", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05979", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "governance", "evaluation_metrics"]}}
{"id": "train_05980", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "governance"]}}
{"id": "train_05981", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tooling"]}}
{"id": "train_05982", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "tooling", "documentation"]}}
{"id": "train_05983", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "tests_are_truth", "governance"]}}
{"id": "train_05984", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05985", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05986", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_05987", "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\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05988", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_05989", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "reproducibility", "tests_are_truth"]}}
{"id": "train_05990", "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\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "security_gates", "documentation"]}}
{"id": "train_05991", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "reproducibility"]}}
{"id": "train_05992", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tests_are_truth", "documentation"]}}
{"id": "train_05993", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "governance", "evaluation_metrics"]}}
{"id": "train_05994", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_05995", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "governance"]}}
{"id": "train_05996", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_05997", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_05998", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "documentation", "repo_scale_reasoning"]}}
{"id": "train_05999", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "repo_scale_reasoning", "governance"]}}
{"id": "train_06000", "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\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "documentation", "governance"]}}
{"id": "train_06001", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "governance"]}}
{"id": "train_06002", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "evaluation_metrics", "reproducibility"]}}
{"id": "train_06003", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_06004", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "ci_integration", "reproducibility"]}}
{"id": "train_06005", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "documentation", "governance"]}}
{"id": "train_06006", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "evaluation_metrics", "documentation"]}}
{"id": "train_06007", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06008", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "documentation"]}}
{"id": "train_06009", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "ci_integration", "reproducibility"]}}
{"id": "train_06010", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "repo_scale_reasoning"]}}
{"id": "train_06011", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_06012", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling"]}}
{"id": "train_06013", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "ci_integration", "tests_are_truth"]}}
{"id": "train_06014", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tooling", "governance"]}}
{"id": "train_06015", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "documentation", "ci_integration"]}}
{"id": "train_06016", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "security_gates"]}}
{"id": "train_06017", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics"]}}
{"id": "train_06018", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "ci_integration", "documentation"]}}
{"id": "train_06019", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "documentation", "ci_integration"]}}
{"id": "train_06020", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "repo_scale_reasoning", "documentation"]}}
{"id": "train_06021", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "documentation", "ci_integration"]}}
{"id": "train_06022", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tests_are_truth", "ci_integration"]}}
{"id": "train_06023", "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\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "tooling"]}}
{"id": "train_06024", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_06025", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_06026", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "governance", "tests_are_truth"]}}
{"id": "train_06027", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "governance", "documentation"]}}
{"id": "train_06028", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_06029", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "ci_integration", "reproducibility"]}}
{"id": "train_06030", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "documentation", "ci_integration"]}}
{"id": "train_06031", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_06032", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "governance", "evaluation_metrics"]}}
{"id": "train_06033", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_06034", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "governance", "evaluation_metrics"]}}
{"id": "train_06035", "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\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_06036", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "documentation"]}}
{"id": "train_06037", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tests_are_truth", "documentation"]}}
{"id": "train_06038", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "documentation", "governance"]}}
{"id": "train_06039", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_06040", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tooling", "governance"]}}
{"id": "train_06041", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "documentation", "ci_integration"]}}
{"id": "train_06042", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "security_gates"]}}
{"id": "train_06043", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_06044", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tests_are_truth", "reproducibility"]}}
{"id": "train_06045", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling"]}}
{"id": "train_06046", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "security_gates", "documentation"]}}
{"id": "train_06047", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "tooling", "ci_integration"]}}
{"id": "train_06048", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "evaluation_metrics", "reproducibility"]}}
{"id": "train_06049", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_06050", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "governance"]}}
{"id": "train_06051", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "security_gates", "tests_are_truth"]}}
{"id": "train_06052", "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\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "evaluation_metrics", "ci_integration"]}}
{"id": "train_06053", "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\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_06054", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_06055", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning"]}}
{"id": "train_06056", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "governance", "repo_scale_reasoning"]}}
{"id": "train_06057", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["governance", "repo_scale_reasoning", "tooling"]}}
{"id": "train_06058", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "reproducibility", "evaluation_metrics"]}}
{"id": "train_06059", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning"]}}
{"id": "train_06060", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "evaluation_metrics", "governance"]}}
{"id": "train_06061", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_06062", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "ci_integration", "tooling"]}}
{"id": "train_06063", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_06064", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility"]}}
{"id": "train_06065", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "tests_are_truth", "security_gates"]}}
{"id": "train_06066", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "reproducibility", "evaluation_metrics"]}}
{"id": "train_06067", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_06068", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "governance"]}}
{"id": "train_06069", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "ci_integration"]}}
{"id": "train_06070", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration"]}}
{"id": "train_06071", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "evaluation_metrics", "ci_integration"]}}
{"id": "train_06072", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility"]}}
{"id": "train_06073", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "documentation", "reproducibility"]}}
{"id": "train_06074", "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\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "reproducibility", "security_gates"]}}
{"id": "train_06075", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling"]}}
{"id": "train_06076", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tests_are_truth", "documentation"]}}
{"id": "train_06077", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_06078", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "ci_integration", "documentation"]}}
{"id": "train_06079", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "reproducibility", "documentation"]}}
{"id": "train_06080", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tests_are_truth", "documentation"]}}
{"id": "train_06081", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tests_are_truth", "reproducibility"]}}
{"id": "train_06082", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_06083", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_06084", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "repo_scale_reasoning", "documentation"]}}
{"id": "train_06085", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration"]}}
{"id": "train_06086", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_06087", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_06088", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_06089", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_06090", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "reproducibility", "ci_integration"]}}
{"id": "train_06091", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "tests_are_truth"]}}
{"id": "train_06092", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "documentation", "security_gates"]}}
{"id": "train_06093", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "reproducibility", "governance"]}}
{"id": "train_06094", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tests_are_truth", "reproducibility"]}}
{"id": "train_06095", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "reproducibility", "ci_integration"]}}
{"id": "train_06096", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_06097", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_06098", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_06099", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "reproducibility", "governance"]}}
{"id": "train_06100", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "evaluation_metrics"]}}
{"id": "train_06101", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "governance", "repo_scale_reasoning"]}}
{"id": "train_06102", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "evaluation_metrics", "ci_integration"]}}
{"id": "train_06103", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "ci_integration", "security_gates"]}}
{"id": "train_06104", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_06105", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "ci_integration", "evaluation_metrics"]}}
{"id": "train_06106", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tests_are_truth", "documentation"]}}
{"id": "train_06107", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "governance"]}}
{"id": "train_06108", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_06109", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "reproducibility", "security_gates"]}}
{"id": "train_06110", "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\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_06111", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "security_gates", "tests_are_truth"]}}
{"id": "train_06112", "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\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "tests_are_truth", "security_gates"]}}
{"id": "train_06113", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_06114", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tooling", "ci_integration"]}}
{"id": "train_06115", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "tooling", "tests_are_truth"]}}
{"id": "train_06116", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "documentation"]}}
{"id": "train_06117", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "security_gates", "reproducibility"]}}
{"id": "train_06118", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics"]}}
{"id": "train_06119", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_06120", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_06121", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_06122", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "documentation", "tooling"]}}
{"id": "train_06123", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "documentation"]}}
{"id": "train_06124", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_06125", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_06126", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "governance", "tooling"]}}
{"id": "train_06127", "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\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_06128", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_06129", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "governance", "documentation"]}}
{"id": "train_06130", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_06131", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_06132", "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\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_06133", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "security_gates", "governance"]}}
{"id": "train_06134", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "reproducibility"]}}
{"id": "train_06135", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "ci_integration", "tooling"]}}
{"id": "train_06136", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "security_gates", "governance"]}}
{"id": "train_06137", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tooling", "reproducibility"]}}
{"id": "train_06138", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_06139", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_06140", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_06141", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "documentation", "reproducibility"]}}
{"id": "train_06142", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_06143", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "evaluation_metrics", "governance"]}}
{"id": "train_06144", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "ci_integration"]}}
{"id": "train_06145", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "reproducibility"]}}
{"id": "train_06146", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_06147", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_06148", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "tooling"]}}
{"id": "train_06149", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "governance", "reproducibility"]}}
{"id": "train_06150", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "reproducibility", "tooling"]}}
{"id": "train_06151", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "governance", "reproducibility"]}}
{"id": "train_06152", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "documentation", "repo_scale_reasoning"]}}
{"id": "train_06153", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "documentation", "tooling"]}}
{"id": "train_06154", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_06155", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "ci_integration"]}}
{"id": "train_06156", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_06157", "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\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06158", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tests_are_truth", "documentation"]}}
{"id": "train_06159", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling"]}}
{"id": "train_06160", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "ci_integration", "security_gates"]}}
{"id": "train_06161", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "governance", "security_gates"]}}
{"id": "train_06162", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "security_gates", "governance"]}}
{"id": "train_06163", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates"]}}
{"id": "train_06164", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "evaluation_metrics", "tooling"]}}
{"id": "train_06165", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_06166", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_06167", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tests_are_truth", "tooling"]}}
{"id": "train_06168", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tooling", "evaluation_metrics"]}}
{"id": "train_06169", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_06170", "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\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning"]}}
{"id": "train_06171", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_06172", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility"]}}
{"id": "train_06173", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "governance", "tooling"]}}
{"id": "train_06174", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tests_are_truth", "tooling"]}}
{"id": "train_06175", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "reproducibility"]}}
{"id": "train_06176", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "repo_scale_reasoning", "documentation"]}}
{"id": "train_06177", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "tooling", "ci_integration"]}}
{"id": "train_06178", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_06179", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_06180", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tests_are_truth", "governance"]}}
{"id": "train_06181", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "ci_integration", "tests_are_truth"]}}
{"id": "train_06182", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "reproducibility", "governance"]}}
{"id": "train_06183", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning"]}}
{"id": "train_06184", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tooling", "security_gates"]}}
{"id": "train_06185", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_06186", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06187", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "ci_integration", "documentation"]}}
{"id": "train_06188", "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\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_06189", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "evaluation_metrics", "documentation"]}}
{"id": "train_06190", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "governance", "tooling"]}}
{"id": "train_06191", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_06192", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth"]}}
{"id": "train_06193", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "security_gates"]}}
{"id": "train_06194", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation"]}}
{"id": "train_06195", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "reproducibility", "ci_integration"]}}
{"id": "train_06196", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_06197", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "security_gates"]}}
{"id": "train_06198", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tooling", "tests_are_truth"]}}
{"id": "train_06199", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tooling", "governance"]}}
{"id": "train_06200", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tests_are_truth", "documentation"]}}
{"id": "train_06201", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "reproducibility", "tests_are_truth"]}}
{"id": "train_06202", "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\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_06203", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "ci_integration", "evaluation_metrics"]}}
{"id": "train_06204", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_06205", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "documentation"]}}
{"id": "train_06206", "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\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_06207", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "security_gates", "tooling"]}}
{"id": "train_06208", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "repo_scale_reasoning"]}}
{"id": "train_06209", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility"]}}
{"id": "train_06210", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "security_gates"]}}
{"id": "train_06211", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tests_are_truth", "governance"]}}
{"id": "train_06212", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "documentation", "governance"]}}
{"id": "train_06213", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "governance", "security_gates"]}}
{"id": "train_06214", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "documentation"]}}
{"id": "train_06215", "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\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "ci_integration", "reproducibility"]}}
{"id": "train_06216", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "governance"]}}
{"id": "train_06217", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_06218", "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\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tooling", "tests_are_truth"]}}
{"id": "train_06219", "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\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tests_are_truth", "governance"]}}
{"id": "train_06220", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_06221", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tests_are_truth"]}}
{"id": "train_06222", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "documentation"]}}
{"id": "train_06223", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "documentation", "ci_integration"]}}
{"id": "train_06224", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tooling", "tests_are_truth"]}}
{"id": "train_06225", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_06226", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_06227", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "reproducibility", "ci_integration"]}}
{"id": "train_06228", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "governance", "tests_are_truth"]}}
{"id": "train_06229", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_06230", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "reproducibility"]}}
{"id": "train_06231", "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\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "tests_are_truth", "security_gates"]}}
{"id": "train_06232", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "documentation", "security_gates"]}}
{"id": "train_06233", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tooling", "governance"]}}
{"id": "train_06234", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "security_gates", "documentation"]}}
{"id": "train_06235", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "security_gates", "tests_are_truth"]}}
{"id": "train_06236", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "ci_integration", "governance"]}}
{"id": "train_06237", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "evaluation_metrics"]}}
{"id": "train_06238", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "tooling", "security_gates"]}}
{"id": "train_06239", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_06240", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tooling", "repo_scale_reasoning"]}}
{"id": "train_06241", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_06242", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "reproducibility"]}}
{"id": "train_06243", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_06244", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "ci_integration", "reproducibility"]}}
{"id": "train_06245", "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\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "governance", "documentation"]}}
{"id": "train_06246", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "ci_integration", "documentation"]}}
{"id": "train_06247", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tooling"]}}
{"id": "train_06248", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_06249", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_06250", "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\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tests_are_truth", "tooling"]}}
{"id": "train_06251", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_06252", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_06253", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tests_are_truth", "documentation"]}}
{"id": "train_06254", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tooling", "ci_integration"]}}
{"id": "train_06255", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_06256", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "ci_integration", "documentation"]}}
{"id": "train_06257", "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\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_06258", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "ci_integration"]}}
{"id": "train_06259", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "repo_scale_reasoning"]}}
{"id": "train_06260", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "evaluation_metrics", "reproducibility"]}}
{"id": "train_06261", "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\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration"]}}
{"id": "train_06262", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_06263", "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\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "governance"]}}
{"id": "train_06264", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "documentation", "tests_are_truth"]}}
{"id": "train_06265", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_06266", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_06267", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "documentation", "security_gates"]}}
{"id": "train_06268", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "security_gates", "governance"]}}
{"id": "train_06269", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_06270", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "evaluation_metrics", "reproducibility"]}}
{"id": "train_06271", "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\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "tests_are_truth"]}}
{"id": "train_06272", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "governance", "evaluation_metrics"]}}
{"id": "train_06273", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth"]}}
{"id": "train_06274", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["governance", "tooling", "evaluation_metrics"]}}
{"id": "train_06275", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_06276", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "documentation"]}}
{"id": "train_06277", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tests_are_truth", "documentation"]}}
{"id": "train_06278", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "governance", "documentation"]}}
{"id": "train_06279", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_06280", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_06281", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "reproducibility", "ci_integration"]}}
{"id": "train_06282", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "documentation", "security_gates"]}}
{"id": "train_06283", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates"]}}
{"id": "train_06284", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "documentation", "evaluation_metrics"]}}
{"id": "train_06285", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tests_are_truth", "ci_integration"]}}
{"id": "train_06286", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_06287", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_06288", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "tooling", "reproducibility"]}}
{"id": "train_06289", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "security_gates"]}}
{"id": "train_06290", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "reproducibility", "security_gates"]}}
{"id": "train_06291", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "security_gates"]}}
{"id": "train_06292", "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\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "governance", "repo_scale_reasoning"]}}
{"id": "train_06293", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "governance", "reproducibility"]}}
{"id": "train_06294", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_06295", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_06296", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "ci_integration", "reproducibility"]}}
{"id": "train_06297", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth"]}}
{"id": "train_06298", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_06299", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tooling", "evaluation_metrics"]}}
{"id": "train_06300", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "ci_integration", "governance"]}}
{"id": "train_06301", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "ci_integration", "tooling"]}}
{"id": "train_06302", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_06303", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "documentation"]}}
{"id": "train_06304", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_06305", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "ci_integration"]}}
{"id": "train_06306", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "security_gates", "reproducibility"]}}
{"id": "train_06307", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_06308", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "reproducibility", "ci_integration"]}}
{"id": "train_06309", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_06310", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tooling", "tests_are_truth"]}}
{"id": "train_06311", "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\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "documentation", "governance"]}}
{"id": "train_06312", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "documentation", "evaluation_metrics"]}}
{"id": "train_06313", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "documentation", "ci_integration"]}}
{"id": "train_06314", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tooling", "security_gates"]}}
{"id": "train_06315", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tests_are_truth", "tooling"]}}
{"id": "train_06316", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_06317", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tooling", "reproducibility"]}}
{"id": "train_06318", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_06319", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "tooling", "ci_integration"]}}
{"id": "train_06320", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "tests_are_truth", "reproducibility"]}}
{"id": "train_06321", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "ci_integration"]}}
{"id": "train_06322", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_06323", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth"]}}
{"id": "train_06324", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "documentation", "reproducibility"]}}
{"id": "train_06325", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_06326", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "reproducibility", "tooling"]}}
{"id": "train_06327", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "ci_integration", "reproducibility"]}}
{"id": "train_06328", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "security_gates"]}}
{"id": "train_06329", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_06330", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "reproducibility", "evaluation_metrics"]}}
{"id": "train_06331", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_06332", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_06333", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "governance", "evaluation_metrics"]}}
{"id": "train_06334", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning"]}}
{"id": "train_06335", "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\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tooling", "evaluation_metrics"]}}
{"id": "train_06336", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "tooling", "security_gates"]}}
{"id": "train_06337", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates"]}}
{"id": "train_06338", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "documentation", "tests_are_truth"]}}
{"id": "train_06339", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_06340", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "repo_scale_reasoning", "governance"]}}
{"id": "train_06341", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "documentation", "governance"]}}
{"id": "train_06342", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "reproducibility"]}}
{"id": "train_06343", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_06344", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "tooling", "reproducibility"]}}
{"id": "train_06345", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "reproducibility"]}}
{"id": "train_06346", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "ci_integration", "tooling"]}}
{"id": "train_06347", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "security_gates", "tests_are_truth"]}}
{"id": "train_06348", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "tooling", "ci_integration"]}}
{"id": "train_06349", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "governance"]}}
{"id": "train_06350", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "documentation"]}}
{"id": "train_06351", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_06352", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "repo_scale_reasoning", "governance"]}}
{"id": "train_06353", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_06354", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility"]}}
{"id": "train_06355", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "governance", "ci_integration"]}}
{"id": "train_06356", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tooling", "tests_are_truth"]}}
{"id": "train_06357", "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\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "reproducibility", "governance"]}}
{"id": "train_06358", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "governance", "tooling"]}}
{"id": "train_06359", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "security_gates", "governance"]}}
{"id": "train_06360", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_06361", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tests_are_truth", "documentation"]}}
{"id": "train_06362", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "ci_integration", "security_gates"]}}
{"id": "train_06363", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_06364", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "security_gates", "tests_are_truth"]}}
{"id": "train_06365", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_06366", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "evaluation_metrics", "reproducibility"]}}
{"id": "train_06367", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_06368", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "documentation", "reproducibility"]}}
{"id": "train_06369", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling"]}}
{"id": "train_06370", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "documentation", "evaluation_metrics"]}}
{"id": "train_06371", "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\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "documentation", "tooling"]}}
{"id": "train_06372", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "ci_integration"]}}
{"id": "train_06373", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "evaluation_metrics", "security_gates"]}}
{"id": "train_06374", "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\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tooling", "security_gates"]}}
{"id": "train_06375", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_06376", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_06377", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling"]}}
{"id": "train_06378", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance"]}}
{"id": "train_06379", "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\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_06380", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_06381", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation"]}}
{"id": "train_06382", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "security_gates", "governance"]}}
{"id": "train_06383", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_06384", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_06385", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_06386", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "evaluation_metrics", "governance"]}}
{"id": "train_06387", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_06388", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "tests_are_truth"]}}
{"id": "train_06389", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_06390", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates"]}}
{"id": "train_06391", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_06392", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "documentation", "tooling"]}}
{"id": "train_06393", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "documentation"]}}
{"id": "train_06394", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tests_are_truth", "governance"]}}
{"id": "train_06395", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "governance"]}}
{"id": "train_06396", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_06397", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_06398", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tooling", "evaluation_metrics"]}}
{"id": "train_06399", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "tests_are_truth", "documentation"]}}
{"id": "train_06400", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_06401", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "repo_scale_reasoning", "documentation"]}}
{"id": "train_06402", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "governance", "tooling"]}}
{"id": "train_06403", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tooling", "evaluation_metrics"]}}
{"id": "train_06404", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_06405", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_06406", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_06407", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "documentation", "ci_integration"]}}
{"id": "train_06408", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_06409", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tooling"]}}
{"id": "train_06410", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "reproducibility"]}}
{"id": "train_06411", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "governance", "security_gates"]}}
{"id": "train_06412", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "governance", "evaluation_metrics"]}}
{"id": "train_06413", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tooling", "governance"]}}
{"id": "train_06414", "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\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_06415", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_06416", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "ci_integration"]}}
{"id": "train_06417", "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\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "reproducibility", "documentation"]}}
{"id": "train_06418", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_06419", "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\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "governance", "ci_integration"]}}
{"id": "train_06420", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "reproducibility", "ci_integration"]}}
{"id": "train_06421", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["governance", "reproducibility", "evaluation_metrics"]}}
{"id": "train_06422", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "documentation"]}}
{"id": "train_06423", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "security_gates", "evaluation_metrics"]}}
{"id": "train_06424", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06425", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "security_gates", "evaluation_metrics"]}}
{"id": "train_06426", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "tooling"]}}
{"id": "train_06427", "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\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06428", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_06429", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tooling", "governance"]}}
{"id": "train_06430", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_06431", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "documentation", "evaluation_metrics"]}}
{"id": "train_06432", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "governance", "repo_scale_reasoning"]}}
{"id": "train_06433", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "evaluation_metrics", "ci_integration"]}}
{"id": "train_06434", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "documentation", "reproducibility"]}}
{"id": "train_06435", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "repo_scale_reasoning", "documentation"]}}
{"id": "train_06436", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_06437", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_06438", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tooling", "governance"]}}
{"id": "train_06439", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_06440", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_06441", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "governance", "ci_integration"]}}
{"id": "train_06442", "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\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_06443", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tests_are_truth", "security_gates"]}}
{"id": "train_06444", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "documentation", "tests_are_truth"]}}
{"id": "train_06445", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "tests_are_truth", "security_gates"]}}
{"id": "train_06446", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "reproducibility"]}}
{"id": "train_06447", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "documentation", "repo_scale_reasoning"]}}
{"id": "train_06448", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "reproducibility", "ci_integration"]}}
{"id": "train_06449", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance"]}}
{"id": "train_06450", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_06451", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "security_gates", "documentation"]}}
{"id": "train_06452", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "documentation", "governance"]}}
{"id": "train_06453", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tooling"]}}
{"id": "train_06454", "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\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "governance", "security_gates"]}}
{"id": "train_06455", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "governance"]}}
{"id": "train_06456", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "ci_integration", "documentation"]}}
{"id": "train_06457", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_06458", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "ci_integration", "tests_are_truth"]}}
{"id": "train_06459", "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\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_06460", "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\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "security_gates"]}}
{"id": "train_06461", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_06462", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "documentation"]}}
{"id": "train_06463", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06464", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "reproducibility", "governance"]}}
{"id": "train_06465", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_06466", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_06467", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_06468", "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\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "security_gates", "governance"]}}
{"id": "train_06469", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_06470", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "governance", "documentation"]}}
{"id": "train_06471", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "ci_integration", "tooling"]}}
{"id": "train_06472", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "security_gates"]}}
{"id": "train_06473", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_06474", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "evaluation_metrics", "ci_integration"]}}
{"id": "train_06475", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates"]}}
{"id": "train_06476", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "governance", "repo_scale_reasoning"]}}
{"id": "train_06477", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning"]}}
{"id": "train_06478", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_06479", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tests_are_truth", "reproducibility"]}}
{"id": "train_06480", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "tooling"]}}
{"id": "train_06481", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tooling"]}}
{"id": "train_06482", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tooling", "security_gates"]}}
{"id": "train_06483", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tests_are_truth", "governance"]}}
{"id": "train_06484", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tooling", "repo_scale_reasoning"]}}
{"id": "train_06485", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_06486", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "ci_integration", "tooling"]}}
{"id": "train_06487", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tooling"]}}
{"id": "train_06488", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "documentation", "ci_integration"]}}
{"id": "train_06489", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "evaluation_metrics"]}}
{"id": "train_06490", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates"]}}
{"id": "train_06491", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_06492", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_06493", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "evaluation_metrics", "security_gates"]}}
{"id": "train_06494", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "reproducibility"]}}
{"id": "train_06495", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "reproducibility", "documentation"]}}
{"id": "train_06496", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_06497", "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\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "documentation"]}}
{"id": "train_06498", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_06499", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "governance", "repo_scale_reasoning"]}}
{"id": "train_06500", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tooling", "documentation"]}}
{"id": "train_06501", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation"]}}
{"id": "train_06502", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "documentation", "governance"]}}
{"id": "train_06503", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "governance", "repo_scale_reasoning"]}}
{"id": "train_06504", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "documentation"]}}
{"id": "train_06505", "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\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_06506", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_06507", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06508", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "ci_integration", "documentation"]}}
{"id": "train_06509", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "governance", "reproducibility"]}}
{"id": "train_06510", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "governance", "ci_integration"]}}
{"id": "train_06511", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_06512", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "reproducibility"]}}
{"id": "train_06513", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "security_gates"]}}
{"id": "train_06514", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tooling", "governance"]}}
{"id": "train_06515", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_06516", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_06517", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "reproducibility", "security_gates"]}}
{"id": "train_06518", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tooling", "repo_scale_reasoning"]}}
{"id": "train_06519", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_06520", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates"]}}
{"id": "train_06521", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "documentation", "tests_are_truth"]}}
{"id": "train_06522", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_06523", "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\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tooling"]}}
{"id": "train_06524", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_06525", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "tooling", "repo_scale_reasoning"]}}
{"id": "train_06526", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "documentation", "tooling"]}}
{"id": "train_06527", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tooling", "documentation"]}}
{"id": "train_06528", "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\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "reproducibility", "ci_integration"]}}
{"id": "train_06529", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tooling"]}}
{"id": "train_06530", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_06531", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance"]}}
{"id": "train_06532", "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\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "reproducibility"]}}
{"id": "train_06533", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_06534", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "documentation", "repo_scale_reasoning"]}}
{"id": "train_06535", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_06536", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "evaluation_metrics", "reproducibility"]}}
{"id": "train_06537", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "documentation", "reproducibility"]}}
{"id": "train_06538", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_06539", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "repo_scale_reasoning", "governance"]}}
{"id": "train_06540", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_06541", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_06542", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_06543", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "governance", "tests_are_truth"]}}
{"id": "train_06544", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "documentation", "repo_scale_reasoning"]}}
{"id": "train_06545", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "ci_integration", "tooling"]}}
{"id": "train_06546", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tooling", "governance"]}}
{"id": "train_06547", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "documentation", "tooling"]}}
{"id": "train_06548", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06549", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_06550", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_06551", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "governance"]}}
{"id": "train_06552", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "governance", "evaluation_metrics"]}}
{"id": "train_06553", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tooling", "tests_are_truth"]}}
{"id": "train_06554", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_06555", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_06556", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tests_are_truth", "tooling"]}}
{"id": "train_06557", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth"]}}
{"id": "train_06558", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "governance", "documentation"]}}
{"id": "train_06559", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_06560", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_06561", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_06562", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "reproducibility", "ci_integration"]}}
{"id": "train_06563", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06564", "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\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tooling", "evaluation_metrics"]}}
{"id": "train_06565", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_06566", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "governance", "repo_scale_reasoning"]}}
{"id": "train_06567", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_06568", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "documentation", "tooling"]}}
{"id": "train_06569", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "security_gates", "evaluation_metrics"]}}
{"id": "train_06570", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tooling", "tests_are_truth"]}}
{"id": "train_06571", "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\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "evaluation_metrics", "ci_integration"]}}
{"id": "train_06572", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "documentation", "security_gates"]}}
{"id": "train_06573", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "governance", "security_gates"]}}
{"id": "train_06574", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_06575", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "reproducibility"]}}
{"id": "train_06576", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_06577", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_06578", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_06579", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "security_gates", "tests_are_truth"]}}
{"id": "train_06580", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "governance"]}}
{"id": "train_06581", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_06582", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "reproducibility", "governance"]}}
{"id": "train_06583", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "tooling"]}}
{"id": "train_06584", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "repo_scale_reasoning"]}}
{"id": "train_06585", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_06586", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_06587", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "documentation", "security_gates"]}}
{"id": "train_06588", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_06589", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_06590", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "tooling"]}}
{"id": "train_06591", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "documentation", "governance"]}}
{"id": "train_06592", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "reproducibility", "evaluation_metrics"]}}
{"id": "train_06593", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tests_are_truth"]}}
{"id": "train_06594", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "security_gates", "ci_integration"]}}
{"id": "train_06595", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_06596", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_06597", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_06598", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "documentation", "governance"]}}
{"id": "train_06599", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06600", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "reproducibility", "tooling"]}}
{"id": "train_06601", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "documentation", "tooling"]}}
{"id": "train_06602", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "governance", "documentation"]}}
{"id": "train_06603", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "governance", "tooling"]}}
{"id": "train_06604", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "documentation", "repo_scale_reasoning"]}}
{"id": "train_06605", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06606", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06607", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "ci_integration", "security_gates"]}}
{"id": "train_06608", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tooling"]}}
{"id": "train_06609", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tests_are_truth", "reproducibility"]}}
{"id": "train_06610", "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\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_06611", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_06612", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_06613", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "ci_integration", "documentation"]}}
{"id": "train_06614", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_06615", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "governance", "evaluation_metrics"]}}
{"id": "train_06616", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "documentation"]}}
{"id": "train_06617", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling"]}}
{"id": "train_06618", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "tests_are_truth", "reproducibility"]}}
{"id": "train_06619", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "evaluation_metrics", "ci_integration"]}}
{"id": "train_06620", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_06621", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "documentation", "tests_are_truth"]}}
{"id": "train_06622", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_06623", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_06624", "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\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_06625", "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\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_06626", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "tooling"]}}
{"id": "train_06627", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_06628", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "governance"]}}
{"id": "train_06629", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "governance"]}}
{"id": "train_06630", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance"]}}
{"id": "train_06631", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "reproducibility", "ci_integration"]}}
{"id": "train_06632", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_06633", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "governance"]}}
{"id": "train_06634", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_06635", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "reproducibility", "evaluation_metrics"]}}
{"id": "train_06636", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "governance"]}}
{"id": "train_06637", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "ci_integration", "governance"]}}
{"id": "train_06638", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "ci_integration"]}}
{"id": "train_06639", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "ci_integration"]}}
{"id": "train_06640", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_06641", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "reproducibility", "governance"]}}
{"id": "train_06642", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics"]}}
{"id": "train_06643", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "security_gates"]}}
{"id": "train_06644", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "ci_integration", "tests_are_truth"]}}
{"id": "train_06645", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_06646", "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\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "security_gates"]}}
{"id": "train_06647", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_06648", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "documentation", "governance"]}}
{"id": "train_06649", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "tests_are_truth"]}}
{"id": "train_06650", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_06651", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_06652", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation"]}}
{"id": "train_06653", "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\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_06654", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_06655", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "tooling", "repo_scale_reasoning"]}}
{"id": "train_06656", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "security_gates"]}}
{"id": "train_06657", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_06658", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tooling", "governance"]}}
{"id": "train_06659", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tooling"]}}
{"id": "train_06660", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_06661", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "evaluation_metrics", "tooling"]}}
{"id": "train_06662", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_06663", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tests_are_truth", "documentation"]}}
{"id": "train_06664", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "tooling", "reproducibility"]}}
{"id": "train_06665", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06666", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_06667", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "governance"]}}
{"id": "train_06668", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_06669", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "reproducibility", "tooling"]}}
{"id": "train_06670", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06671", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "security_gates"]}}
{"id": "train_06672", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "governance", "documentation"]}}
{"id": "train_06673", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "repo_scale_reasoning", "governance"]}}
{"id": "train_06674", "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\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_06675", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tooling", "ci_integration"]}}
{"id": "train_06676", "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\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_06677", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tooling"]}}
{"id": "train_06678", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "reproducibility", "ci_integration"]}}
{"id": "train_06679", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "reproducibility", "documentation"]}}
{"id": "train_06680", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_06681", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "reproducibility", "tests_are_truth"]}}
{"id": "train_06682", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "security_gates", "reproducibility"]}}
{"id": "train_06683", "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\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "documentation"]}}
{"id": "train_06684", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "reproducibility"]}}
{"id": "train_06685", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_06686", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "tooling", "evaluation_metrics"]}}
{"id": "train_06687", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tooling", "repo_scale_reasoning"]}}
{"id": "train_06688", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "evaluation_metrics", "documentation"]}}
{"id": "train_06689", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance"]}}
{"id": "train_06690", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_06691", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "ci_integration"]}}
{"id": "train_06692", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "evaluation_metrics"]}}
{"id": "train_06693", "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\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "governance"]}}
{"id": "train_06694", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tooling", "tests_are_truth"]}}
{"id": "train_06695", "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\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "evaluation_metrics", "security_gates"]}}
{"id": "train_06696", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "reproducibility", "evaluation_metrics"]}}
{"id": "train_06697", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "documentation", "governance"]}}
{"id": "train_06698", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "ci_integration", "documentation"]}}
{"id": "train_06699", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_06700", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "repo_scale_reasoning", "governance"]}}
{"id": "train_06701", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "ci_integration"]}}
{"id": "train_06702", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "security_gates"]}}
{"id": "train_06703", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "documentation"]}}
{"id": "train_06704", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "security_gates", "documentation"]}}
{"id": "train_06705", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "ci_integration", "tests_are_truth"]}}
{"id": "train_06706", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "documentation", "evaluation_metrics"]}}
{"id": "train_06707", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "security_gates", "reproducibility"]}}
{"id": "train_06708", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "governance", "repo_scale_reasoning"]}}
{"id": "train_06709", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "ci_integration", "governance"]}}
{"id": "train_06710", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "documentation", "ci_integration"]}}
{"id": "train_06711", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_06712", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "documentation", "tooling"]}}
{"id": "train_06713", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_06714", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tooling"]}}
{"id": "train_06715", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "security_gates", "documentation"]}}
{"id": "train_06716", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tooling", "tests_are_truth"]}}
{"id": "train_06717", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "documentation"]}}
{"id": "train_06718", "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\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "evaluation_metrics"]}}
{"id": "train_06719", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "tooling"]}}
{"id": "train_06720", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_06721", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility"]}}
{"id": "train_06722", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_06723", "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\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "tests_are_truth", "reproducibility"]}}
{"id": "train_06724", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_06725", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tests_are_truth"]}}
{"id": "train_06726", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "evaluation_metrics", "ci_integration"]}}
{"id": "train_06727", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "governance", "tooling"]}}
{"id": "train_06728", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tooling", "documentation"]}}
{"id": "train_06729", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tooling", "reproducibility"]}}
{"id": "train_06730", "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\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_06731", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "governance", "tests_are_truth"]}}
{"id": "train_06732", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "security_gates", "documentation"]}}
{"id": "train_06733", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tooling"]}}
{"id": "train_06734", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "security_gates", "tooling"]}}
{"id": "train_06735", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "security_gates", "documentation"]}}
{"id": "train_06736", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "repo_scale_reasoning", "tooling"]}}
{"id": "train_06737", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tooling", "documentation"]}}
{"id": "train_06738", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "reproducibility", "documentation"]}}
{"id": "train_06739", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "evaluation_metrics", "governance"]}}
{"id": "train_06740", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_06741", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "ci_integration", "evaluation_metrics"]}}
{"id": "train_06742", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance"]}}
{"id": "train_06743", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "evaluation_metrics"]}}
{"id": "train_06744", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_06745", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tooling", "governance"]}}
{"id": "train_06746", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "tooling", "reproducibility"]}}
{"id": "train_06747", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "reproducibility"]}}
{"id": "train_06748", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "documentation"]}}
{"id": "train_06749", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_06750", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_06751", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "reproducibility"]}}
{"id": "train_06752", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_06753", "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\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_06754", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "security_gates"]}}
{"id": "train_06755", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "security_gates", "ci_integration"]}}
{"id": "train_06756", "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\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "ci_integration"]}}
{"id": "train_06757", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "reproducibility", "governance"]}}
{"id": "train_06758", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_06759", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tooling"]}}
{"id": "train_06760", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics"]}}
{"id": "train_06761", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_06762", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "documentation", "repo_scale_reasoning"]}}
{"id": "train_06763", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_06764", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "governance"]}}
{"id": "train_06765", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "tooling"]}}
{"id": "train_06766", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "governance", "evaluation_metrics"]}}
{"id": "train_06767", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "security_gates", "tooling"]}}
{"id": "train_06768", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_06769", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "ci_integration", "tests_are_truth"]}}
{"id": "train_06770", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_06771", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "governance", "documentation"]}}
{"id": "train_06772", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "reproducibility", "security_gates"]}}
{"id": "train_06773", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_06774", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_06775", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_06776", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "governance", "security_gates"]}}
{"id": "train_06777", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tooling", "documentation"]}}
{"id": "train_06778", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "evaluation_metrics", "tooling"]}}
{"id": "train_06779", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tooling"]}}
{"id": "train_06780", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "documentation", "tooling"]}}
{"id": "train_06781", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_06782", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "governance", "tooling"]}}
{"id": "train_06783", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "evaluation_metrics", "ci_integration"]}}
{"id": "train_06784", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tooling", "evaluation_metrics"]}}
{"id": "train_06785", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "documentation", "tests_are_truth"]}}
{"id": "train_06786", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "documentation", "reproducibility"]}}
{"id": "train_06787", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "reproducibility"]}}
{"id": "train_06788", "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\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "documentation", "tooling"]}}
{"id": "train_06789", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tests_are_truth", "reproducibility"]}}
{"id": "train_06790", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_06791", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tests_are_truth", "reproducibility"]}}
{"id": "train_06792", "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\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_06793", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "ci_integration", "governance"]}}
{"id": "train_06794", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tests_are_truth", "reproducibility"]}}
{"id": "train_06795", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "reproducibility", "ci_integration"]}}
{"id": "train_06796", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration"]}}
{"id": "train_06797", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "documentation", "tests_are_truth"]}}
{"id": "train_06798", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "reproducibility", "security_gates"]}}
{"id": "train_06799", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "evaluation_metrics", "security_gates"]}}
{"id": "train_06800", "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\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance"]}}
{"id": "train_06801", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06802", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_06803", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_06804", "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\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_06805", "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\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "documentation", "reproducibility"]}}
{"id": "train_06806", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_06807", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "repo_scale_reasoning"]}}
{"id": "train_06808", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_06809", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "security_gates"]}}
{"id": "train_06810", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_06811", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "documentation", "tests_are_truth"]}}
{"id": "train_06812", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "documentation"]}}
{"id": "train_06813", "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\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "evaluation_metrics", "tooling"]}}
{"id": "train_06814", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tests_are_truth", "documentation"]}}
{"id": "train_06815", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "governance"]}}
{"id": "train_06816", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_06817", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "documentation", "reproducibility"]}}
{"id": "train_06818", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration"]}}
{"id": "train_06819", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "evaluation_metrics"]}}
{"id": "train_06820", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tooling", "ci_integration"]}}
{"id": "train_06821", "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\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tests_are_truth", "tooling"]}}
{"id": "train_06822", "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\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_06823", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "security_gates"]}}
{"id": "train_06824", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "reproducibility", "documentation"]}}
{"id": "train_06825", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_06826", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_06827", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility"]}}
{"id": "train_06828", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_06829", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "reproducibility", "security_gates"]}}
{"id": "train_06830", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "security_gates"]}}
{"id": "train_06831", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "governance"]}}
{"id": "train_06832", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates"]}}
{"id": "train_06833", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "repo_scale_reasoning"]}}
{"id": "train_06834", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tooling"]}}
{"id": "train_06835", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "reproducibility", "evaluation_metrics"]}}
{"id": "train_06836", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "governance", "reproducibility"]}}
{"id": "train_06837", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_06838", "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\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "security_gates", "ci_integration"]}}
{"id": "train_06839", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "evaluation_metrics"]}}
{"id": "train_06840", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_06841", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "reproducibility"]}}
{"id": "train_06842", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "reproducibility", "governance"]}}
{"id": "train_06843", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "evaluation_metrics"]}}
{"id": "train_06844", "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\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "security_gates", "tests_are_truth"]}}
{"id": "train_06845", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "reproducibility", "documentation"]}}
{"id": "train_06846", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "governance", "security_gates"]}}
{"id": "train_06847", "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\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "documentation", "reproducibility"]}}
{"id": "train_06848", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_06849", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_06850", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tooling", "tests_are_truth"]}}
{"id": "train_06851", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_06852", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "reproducibility", "governance"]}}
{"id": "train_06853", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "evaluation_metrics", "documentation"]}}
{"id": "train_06854", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_06855", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_06856", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_06857", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_06858", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tooling", "reproducibility"]}}
{"id": "train_06859", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_06860", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation"]}}
{"id": "train_06861", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_06862", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "governance"]}}
{"id": "train_06863", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "governance", "repo_scale_reasoning"]}}
{"id": "train_06864", "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\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_06865", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_06866", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "documentation", "security_gates"]}}
{"id": "train_06867", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "ci_integration", "reproducibility"]}}
{"id": "train_06868", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_06869", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "tooling"]}}
{"id": "train_06870", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tooling", "evaluation_metrics"]}}
{"id": "train_06871", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "documentation"]}}
{"id": "train_06872", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_06873", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "governance", "tooling"]}}
{"id": "train_06874", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "ci_integration", "tooling"]}}
{"id": "train_06875", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "reproducibility", "documentation"]}}
{"id": "train_06876", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "documentation", "governance"]}}
{"id": "train_06877", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "documentation", "repo_scale_reasoning"]}}
{"id": "train_06878", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_06879", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_06880", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "tooling"]}}
{"id": "train_06881", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_06882", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_06883", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "security_gates", "documentation"]}}
{"id": "train_06884", "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\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tests_are_truth", "ci_integration"]}}
{"id": "train_06885", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "reproducibility", "security_gates"]}}
{"id": "train_06886", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "governance", "evaluation_metrics"]}}
{"id": "train_06887", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tests_are_truth", "documentation"]}}
{"id": "train_06888", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "governance", "tooling"]}}
{"id": "train_06889", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "governance", "ci_integration"]}}
{"id": "train_06890", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation"]}}
{"id": "train_06891", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "tests_are_truth"]}}
{"id": "train_06892", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "governance", "documentation"]}}
{"id": "train_06893", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "reproducibility", "ci_integration"]}}
{"id": "train_06894", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_06895", "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\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tooling", "tests_are_truth"]}}
{"id": "train_06896", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "documentation", "tests_are_truth"]}}
{"id": "train_06897", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tooling", "repo_scale_reasoning"]}}
{"id": "train_06898", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "ci_integration", "reproducibility"]}}
{"id": "train_06899", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tooling", "evaluation_metrics"]}}
{"id": "train_06900", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_06901", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_06902", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "tooling"]}}
{"id": "train_06903", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "ci_integration", "reproducibility"]}}
{"id": "train_06904", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tests_are_truth", "security_gates"]}}
{"id": "train_06905", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "documentation", "governance"]}}
{"id": "train_06906", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "reproducibility", "documentation"]}}
{"id": "train_06907", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "documentation"]}}
{"id": "train_06908", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tests_are_truth", "documentation"]}}
{"id": "train_06909", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_06910", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tooling", "evaluation_metrics"]}}
{"id": "train_06911", "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\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tooling", "documentation"]}}
{"id": "train_06912", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "security_gates", "reproducibility"]}}
{"id": "train_06913", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "reproducibility", "tooling"]}}
{"id": "train_06914", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "security_gates", "tooling"]}}
{"id": "train_06915", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_06916", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "documentation", "tests_are_truth"]}}
{"id": "train_06917", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "ci_integration", "documentation"]}}
{"id": "train_06918", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_06919", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "ci_integration"]}}
{"id": "train_06920", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tooling"]}}
{"id": "train_06921", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_06922", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tests_are_truth"]}}
{"id": "train_06923", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_06924", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "evaluation_metrics", "governance"]}}
{"id": "train_06925", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "security_gates"]}}
{"id": "train_06926", "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\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "governance", "tooling"]}}
{"id": "train_06927", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_06928", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "governance"]}}
{"id": "train_06929", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "documentation", "reproducibility"]}}
{"id": "train_06930", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_06931", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tests_are_truth", "security_gates"]}}
{"id": "train_06932", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_06933", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tests_are_truth", "governance"]}}
{"id": "train_06934", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_06935", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "documentation", "reproducibility"]}}
{"id": "train_06936", "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\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "reproducibility", "tests_are_truth"]}}
{"id": "train_06937", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "governance", "tooling"]}}
{"id": "train_06938", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_06939", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_06940", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling"]}}
{"id": "train_06941", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "ci_integration"]}}
{"id": "train_06942", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tooling", "evaluation_metrics"]}}
{"id": "train_06943", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth"]}}
{"id": "train_06944", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_06945", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_06946", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "documentation", "ci_integration"]}}
{"id": "train_06947", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_06948", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_06949", "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\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "evaluation_metrics", "governance"]}}
{"id": "train_06950", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_06951", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "security_gates", "tests_are_truth"]}}
{"id": "train_06952", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tooling", "evaluation_metrics"]}}
{"id": "train_06953", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_06954", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics"]}}
{"id": "train_06955", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "evaluation_metrics", "tooling"]}}
{"id": "train_06956", "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\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth"]}}
{"id": "train_06957", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_06958", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "governance"]}}
{"id": "train_06959", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "governance"]}}
{"id": "train_06960", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_06961", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_06962", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_06963", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "evaluation_metrics", "ci_integration"]}}
{"id": "train_06964", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "reproducibility", "documentation"]}}
{"id": "train_06965", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_06966", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "documentation"]}}
{"id": "train_06967", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "documentation", "security_gates"]}}
{"id": "train_06968", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "ci_integration", "security_gates"]}}
{"id": "train_06969", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tooling", "reproducibility"]}}
{"id": "train_06970", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling"]}}
{"id": "train_06971", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_06972", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "ci_integration"]}}
{"id": "train_06973", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "documentation"]}}
{"id": "train_06974", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tooling", "reproducibility"]}}
{"id": "train_06975", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "security_gates", "ci_integration"]}}
{"id": "train_06976", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "evaluation_metrics", "ci_integration"]}}
{"id": "train_06977", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_06978", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics"]}}
{"id": "train_06979", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_06980", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "documentation", "governance"]}}
{"id": "train_06981", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_06982", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "reproducibility"]}}
{"id": "train_06983", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "tests_are_truth", "reproducibility"]}}
{"id": "train_06984", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "reproducibility"]}}
{"id": "train_06985", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "documentation"]}}
{"id": "train_06986", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_06987", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_06988", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "reproducibility", "documentation"]}}
{"id": "train_06989", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "security_gates", "evaluation_metrics"]}}
{"id": "train_06990", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics"]}}
{"id": "train_06991", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_06992", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "governance", "security_gates"]}}
{"id": "train_06993", "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\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "ci_integration", "governance"]}}
{"id": "train_06994", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tests_are_truth", "reproducibility"]}}
{"id": "train_06995", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_06996", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "ci_integration", "documentation"]}}
{"id": "train_06997", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_06998", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "governance", "security_gates"]}}
{"id": "train_06999", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "security_gates", "evaluation_metrics"]}}
{"id": "train_07000", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "governance", "ci_integration"]}}
{"id": "train_07001", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_07002", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tests_are_truth", "documentation"]}}
{"id": "train_07003", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "security_gates", "tooling"]}}
{"id": "train_07004", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "documentation", "security_gates"]}}
{"id": "train_07005", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_07006", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates"]}}
{"id": "train_07007", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "ci_integration", "tests_are_truth"]}}
{"id": "train_07008", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_07009", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "documentation", "security_gates"]}}
{"id": "train_07010", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "security_gates", "documentation"]}}
{"id": "train_07011", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "security_gates", "tooling"]}}
{"id": "train_07012", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "reproducibility", "tests_are_truth"]}}
{"id": "train_07013", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_07014", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "documentation", "tooling"]}}
{"id": "train_07015", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "repo_scale_reasoning"]}}
{"id": "train_07016", "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\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "security_gates", "evaluation_metrics"]}}
{"id": "train_07017", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "security_gates", "tests_are_truth"]}}
{"id": "train_07018", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "tooling", "governance"]}}
{"id": "train_07019", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "governance", "tooling"]}}
{"id": "train_07020", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tooling", "tests_are_truth"]}}
{"id": "train_07021", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics"]}}
{"id": "train_07022", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_07023", "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\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "ci_integration", "tooling"]}}
{"id": "train_07024", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07025", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_07026", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "security_gates"]}}
{"id": "train_07027", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "governance"]}}
{"id": "train_07028", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "governance"]}}
{"id": "train_07029", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "reproducibility", "documentation"]}}
{"id": "train_07030", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_07031", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "documentation"]}}
{"id": "train_07032", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "ci_integration", "tests_are_truth"]}}
{"id": "train_07033", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_07034", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_07035", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tooling", "governance"]}}
{"id": "train_07036", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "security_gates", "governance"]}}
{"id": "train_07037", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_07038", "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\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tests_are_truth", "reproducibility"]}}
{"id": "train_07039", "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\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07040", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "evaluation_metrics"]}}
{"id": "train_07041", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_07042", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_07043", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_07044", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_07045", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_07046", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_07047", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "ci_integration", "reproducibility"]}}
{"id": "train_07048", "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\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "governance", "tooling"]}}
{"id": "train_07049", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "repo_scale_reasoning", "governance"]}}
{"id": "train_07050", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_07051", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_07052", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "governance"]}}
{"id": "train_07053", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_07054", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_07055", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tooling", "repo_scale_reasoning"]}}
{"id": "train_07056", "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\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_07057", "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\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "security_gates"]}}
{"id": "train_07058", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07059", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_07060", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "reproducibility"]}}
{"id": "train_07061", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "repo_scale_reasoning"]}}
{"id": "train_07062", "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\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics"]}}
{"id": "train_07063", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "ci_integration"]}}
{"id": "train_07064", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "documentation", "security_gates"]}}
{"id": "train_07065", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "ci_integration", "security_gates"]}}
{"id": "train_07066", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "documentation"]}}
{"id": "train_07067", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tooling", "documentation"]}}
{"id": "train_07068", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "security_gates", "governance"]}}
{"id": "train_07069", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_07070", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "governance", "tooling"]}}
{"id": "train_07071", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_07072", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "documentation", "reproducibility"]}}
{"id": "train_07073", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_07074", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_07075", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "governance", "ci_integration"]}}
{"id": "train_07076", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "documentation", "governance"]}}
{"id": "train_07077", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "documentation", "ci_integration"]}}
{"id": "train_07078", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "governance"]}}
{"id": "train_07079", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "security_gates", "evaluation_metrics"]}}
{"id": "train_07080", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tests_are_truth", "ci_integration"]}}
{"id": "train_07081", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "ci_integration", "tests_are_truth"]}}
{"id": "train_07082", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "governance", "security_gates"]}}
{"id": "train_07083", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "governance", "reproducibility"]}}
{"id": "train_07084", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_07085", "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\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "tooling", "repo_scale_reasoning"]}}
{"id": "train_07086", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "documentation", "tests_are_truth"]}}
{"id": "train_07087", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "documentation", "ci_integration"]}}
{"id": "train_07088", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "governance"]}}
{"id": "train_07089", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning"]}}
{"id": "train_07090", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "security_gates"]}}
{"id": "train_07091", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "ci_integration", "evaluation_metrics"]}}
{"id": "train_07092", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling"]}}
{"id": "train_07093", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling"]}}
{"id": "train_07094", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "reproducibility", "tests_are_truth"]}}
{"id": "train_07095", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tests_are_truth", "reproducibility"]}}
{"id": "train_07096", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "governance", "tooling"]}}
{"id": "train_07097", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "reproducibility", "governance"]}}
{"id": "train_07098", "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\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "reproducibility", "evaluation_metrics"]}}
{"id": "train_07099", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "documentation", "evaluation_metrics"]}}
{"id": "train_07100", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "evaluation_metrics", "ci_integration"]}}
{"id": "train_07101", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tooling", "tests_are_truth"]}}
{"id": "train_07102", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_07103", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07104", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07105", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_07106", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "security_gates"]}}
{"id": "train_07107", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "ci_integration"]}}
{"id": "train_07108", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_07109", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "governance", "ci_integration"]}}
{"id": "train_07110", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "ci_integration", "reproducibility"]}}
{"id": "train_07111", "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\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_07112", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_07113", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_07114", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility"]}}
{"id": "train_07115", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_07116", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07117", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_07118", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "ci_integration", "governance"]}}
{"id": "train_07119", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_07120", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_07121", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics"]}}
{"id": "train_07122", "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\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "documentation", "repo_scale_reasoning"]}}
{"id": "train_07123", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_07124", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "governance", "ci_integration"]}}
{"id": "train_07125", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "reproducibility"]}}
{"id": "train_07126", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "tooling"]}}
{"id": "train_07127", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_07128", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "security_gates", "governance"]}}
{"id": "train_07129", "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\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "security_gates", "documentation"]}}
{"id": "train_07130", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "evaluation_metrics", "ci_integration"]}}
{"id": "train_07131", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "reproducibility", "evaluation_metrics"]}}
{"id": "train_07132", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "governance"]}}
{"id": "train_07133", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_07134", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_07135", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_07136", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tooling"]}}
{"id": "train_07137", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_07138", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_07139", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tooling", "governance"]}}
{"id": "train_07140", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "security_gates", "tests_are_truth"]}}
{"id": "train_07141", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling"]}}
{"id": "train_07142", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "ci_integration", "governance"]}}
{"id": "train_07143", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tests_are_truth", "documentation"]}}
{"id": "train_07144", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "documentation", "ci_integration"]}}
{"id": "train_07145", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_07146", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "evaluation_metrics", "reproducibility"]}}
{"id": "train_07147", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_07148", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tooling", "tests_are_truth"]}}
{"id": "train_07149", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "evaluation_metrics", "ci_integration"]}}
{"id": "train_07150", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling"]}}
{"id": "train_07151", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "tooling", "tests_are_truth"]}}
{"id": "train_07152", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "security_gates"]}}
{"id": "train_07153", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "governance", "ci_integration"]}}
{"id": "train_07154", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "documentation", "ci_integration"]}}
{"id": "train_07155", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_07156", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_07157", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "documentation"]}}
{"id": "train_07158", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_07159", "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\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "governance"]}}
{"id": "train_07160", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_07161", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tooling", "documentation"]}}
{"id": "train_07162", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_07163", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tooling", "evaluation_metrics"]}}
{"id": "train_07164", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "documentation", "governance"]}}
{"id": "train_07165", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "governance", "reproducibility"]}}
{"id": "train_07166", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tests_are_truth", "documentation"]}}
{"id": "train_07167", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_07168", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "governance", "tests_are_truth"]}}
{"id": "train_07169", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "security_gates", "evaluation_metrics"]}}
{"id": "train_07170", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "tooling"]}}
{"id": "train_07171", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_07172", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "governance", "ci_integration"]}}
{"id": "train_07173", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "documentation", "security_gates"]}}
{"id": "train_07174", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "ci_integration"]}}
{"id": "train_07175", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "tests_are_truth"]}}
{"id": "train_07176", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "reproducibility", "evaluation_metrics"]}}
{"id": "train_07177", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "reproducibility", "tests_are_truth"]}}
{"id": "train_07178", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "security_gates"]}}
{"id": "train_07179", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_07180", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "governance"]}}
{"id": "train_07181", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning"]}}
{"id": "train_07182", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "reproducibility", "tooling"]}}
{"id": "train_07183", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_07184", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_07185", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tooling"]}}
{"id": "train_07186", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility"]}}
{"id": "train_07187", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "evaluation_metrics", "security_gates"]}}
{"id": "train_07188", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_07189", "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\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_07190", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "tooling", "reproducibility"]}}
{"id": "train_07191", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "evaluation_metrics", "tooling"]}}
{"id": "train_07192", "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\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_07193", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_07194", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "reproducibility"]}}
{"id": "train_07195", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "security_gates", "tests_are_truth"]}}
{"id": "train_07196", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tooling", "tests_are_truth"]}}
{"id": "train_07197", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "ci_integration"]}}
{"id": "train_07198", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "governance", "tests_are_truth"]}}
{"id": "train_07199", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "tooling", "security_gates"]}}
{"id": "train_07200", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_07201", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_07202", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "governance", "reproducibility"]}}
{"id": "train_07203", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "documentation", "ci_integration"]}}
{"id": "train_07204", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_07205", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_07206", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07207", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility"]}}
{"id": "train_07208", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "governance", "tooling"]}}
{"id": "train_07209", "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\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_07210", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "tests_are_truth", "tooling"]}}
{"id": "train_07211", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "reproducibility", "tooling"]}}
{"id": "train_07212", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "governance", "ci_integration"]}}
{"id": "train_07213", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_07214", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_07215", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "security_gates"]}}
{"id": "train_07216", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tooling", "security_gates"]}}
{"id": "train_07217", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "evaluation_metrics", "security_gates"]}}
{"id": "train_07218", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance"]}}
{"id": "train_07219", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "evaluation_metrics"]}}
{"id": "train_07220", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "ci_integration", "evaluation_metrics"]}}
{"id": "train_07221", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_07222", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "ci_integration", "tests_are_truth"]}}
{"id": "train_07223", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_07224", "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\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "documentation"]}}
{"id": "train_07225", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "governance"]}}
{"id": "train_07226", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "tooling", "documentation"]}}
{"id": "train_07227", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_07228", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "security_gates"]}}
{"id": "train_07229", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_07230", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "governance", "evaluation_metrics"]}}
{"id": "train_07231", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "governance"]}}
{"id": "train_07232", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "security_gates"]}}
{"id": "train_07233", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning"]}}
{"id": "train_07234", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics"]}}
{"id": "train_07235", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "tooling"]}}
{"id": "train_07236", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "governance", "ci_integration"]}}
{"id": "train_07237", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "reproducibility", "security_gates"]}}
{"id": "train_07238", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "reproducibility", "ci_integration"]}}
{"id": "train_07239", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_07240", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tests_are_truth", "tooling"]}}
{"id": "train_07241", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "reproducibility", "tooling"]}}
{"id": "train_07242", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tooling", "documentation"]}}
{"id": "train_07243", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07244", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "governance", "security_gates"]}}
{"id": "train_07245", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "governance", "reproducibility"]}}
{"id": "train_07246", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "evaluation_metrics"]}}
{"id": "train_07247", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "governance", "documentation"]}}
{"id": "train_07248", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "ci_integration", "documentation"]}}
{"id": "train_07249", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "reproducibility"]}}
{"id": "train_07250", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_07251", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_07252", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_07253", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_07254", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tests_are_truth", "documentation"]}}
{"id": "train_07255", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_07256", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "ci_integration"]}}
{"id": "train_07257", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "tests_are_truth"]}}
{"id": "train_07258", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "tooling"]}}
{"id": "train_07259", "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\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling"]}}
{"id": "train_07260", "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\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "evaluation_metrics", "documentation"]}}
{"id": "train_07261", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_07262", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07263", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth"]}}
{"id": "train_07264", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "documentation", "governance"]}}
{"id": "train_07265", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "ci_integration", "documentation"]}}
{"id": "train_07266", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "security_gates"]}}
{"id": "train_07267", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_07268", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tooling", "repo_scale_reasoning"]}}
{"id": "train_07269", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_07270", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_07271", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_07272", "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\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_07273", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tooling", "documentation"]}}
{"id": "train_07274", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_07275", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tests_are_truth", "tooling"]}}
{"id": "train_07276", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "repo_scale_reasoning", "documentation"]}}
{"id": "train_07277", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tooling"]}}
{"id": "train_07278", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling"]}}
{"id": "train_07279", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "security_gates", "ci_integration"]}}
{"id": "train_07280", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_07281", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tooling", "security_gates"]}}
{"id": "train_07282", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "ci_integration"]}}
{"id": "train_07283", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "ci_integration", "governance"]}}
{"id": "train_07284", "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\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tooling", "tests_are_truth"]}}
{"id": "train_07285", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_07286", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_07287", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "documentation", "repo_scale_reasoning"]}}
{"id": "train_07288", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "documentation", "ci_integration"]}}
{"id": "train_07289", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "reproducibility", "tests_are_truth"]}}
{"id": "train_07290", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tooling", "governance"]}}
{"id": "train_07291", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "reproducibility"]}}
{"id": "train_07292", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "ci_integration", "security_gates"]}}
{"id": "train_07293", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_07294", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tooling", "reproducibility"]}}
{"id": "train_07295", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_07296", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "security_gates", "tooling"]}}
{"id": "train_07297", "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\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_07298", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "documentation"]}}
{"id": "train_07299", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_07300", "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\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_07301", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_07302", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "repo_scale_reasoning", "documentation"]}}
{"id": "train_07303", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07304", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_07305", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "reproducibility", "security_gates"]}}
{"id": "train_07306", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07307", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tooling", "reproducibility"]}}
{"id": "train_07308", "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\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics"]}}
{"id": "train_07309", "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\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_07310", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_07311", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_07312", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "governance", "tests_are_truth"]}}
{"id": "train_07313", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_07314", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_07315", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "security_gates", "ci_integration"]}}
{"id": "train_07316", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "security_gates"]}}
{"id": "train_07317", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_07318", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "repo_scale_reasoning", "documentation"]}}
{"id": "train_07319", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_07320", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "evaluation_metrics"]}}
{"id": "train_07321", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_07322", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tooling", "reproducibility"]}}
{"id": "train_07323", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07324", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_07325", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_07326", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tests_are_truth", "ci_integration"]}}
{"id": "train_07327", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tests_are_truth"]}}
{"id": "train_07328", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "governance", "evaluation_metrics"]}}
{"id": "train_07329", "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\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "documentation", "ci_integration"]}}
{"id": "train_07330", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_07331", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation"]}}
{"id": "train_07332", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_07333", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_07334", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "documentation", "tooling"]}}
{"id": "train_07335", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_07336", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tests_are_truth", "tooling"]}}
{"id": "train_07337", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tooling", "reproducibility"]}}
{"id": "train_07338", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "reproducibility", "governance"]}}
{"id": "train_07339", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07340", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_07341", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_07342", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "security_gates"]}}
{"id": "train_07343", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "ci_integration"]}}
{"id": "train_07344", "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\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "documentation", "evaluation_metrics"]}}
{"id": "train_07345", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_07346", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "ci_integration", "governance"]}}
{"id": "train_07347", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "documentation", "tests_are_truth"]}}
{"id": "train_07348", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07349", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "governance", "tests_are_truth"]}}
{"id": "train_07350", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration"]}}
{"id": "train_07351", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "ci_integration", "reproducibility"]}}
{"id": "train_07352", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_07353", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "ci_integration", "evaluation_metrics"]}}
{"id": "train_07354", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_07355", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_07356", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tests_are_truth", "tooling"]}}
{"id": "train_07357", "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\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "security_gates", "evaluation_metrics"]}}
{"id": "train_07358", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "evaluation_metrics", "tooling"]}}
{"id": "train_07359", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "repo_scale_reasoning"]}}
{"id": "train_07360", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_07361", "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\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "evaluation_metrics"]}}
{"id": "train_07362", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance"]}}
{"id": "train_07363", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "reproducibility"]}}
{"id": "train_07364", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_07365", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_07366", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "reproducibility", "ci_integration"]}}
{"id": "train_07367", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "documentation"]}}
{"id": "train_07368", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_07369", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_07370", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "documentation", "governance"]}}
{"id": "train_07371", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "documentation", "security_gates"]}}
{"id": "train_07372", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_07373", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_07374", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "ci_integration", "governance"]}}
{"id": "train_07375", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration"]}}
{"id": "train_07376", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_07377", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "documentation"]}}
{"id": "train_07378", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_07379", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "documentation", "ci_integration"]}}
{"id": "train_07380", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_07381", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "security_gates"]}}
{"id": "train_07382", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_07383", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "governance", "ci_integration"]}}
{"id": "train_07384", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "governance"]}}
{"id": "train_07385", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "security_gates", "governance"]}}
{"id": "train_07386", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "reproducibility", "tests_are_truth"]}}
{"id": "train_07387", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "governance", "evaluation_metrics"]}}
{"id": "train_07388", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tooling", "reproducibility"]}}
{"id": "train_07389", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth"]}}
{"id": "train_07390", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_07391", "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\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "evaluation_metrics"]}}
{"id": "train_07392", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "governance", "repo_scale_reasoning"]}}
{"id": "train_07393", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_07394", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "reproducibility", "governance"]}}
{"id": "train_07395", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_07396", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tooling"]}}
{"id": "train_07397", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tooling", "repo_scale_reasoning"]}}
{"id": "train_07398", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_07399", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates"]}}
{"id": "train_07400", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_07401", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tooling", "repo_scale_reasoning"]}}
{"id": "train_07402", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tooling", "documentation"]}}
{"id": "train_07403", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_07404", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "documentation", "ci_integration"]}}
{"id": "train_07405", "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\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "governance", "security_gates"]}}
{"id": "train_07406", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration"]}}
{"id": "train_07407", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07408", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_07409", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "reproducibility"]}}
{"id": "train_07410", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "reproducibility", "security_gates"]}}
{"id": "train_07411", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "reproducibility"]}}
{"id": "train_07412", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_07413", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "documentation", "reproducibility"]}}
{"id": "train_07414", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_07415", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_07416", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_07417", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "governance", "evaluation_metrics"]}}
{"id": "train_07418", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07419", "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\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "ci_integration", "governance"]}}
{"id": "train_07420", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tooling"]}}
{"id": "train_07421", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tests_are_truth", "documentation"]}}
{"id": "train_07422", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "evaluation_metrics", "reproducibility"]}}
{"id": "train_07423", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "reproducibility"]}}
{"id": "train_07424", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_07425", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "ci_integration", "tooling"]}}
{"id": "train_07426", "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\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tooling", "documentation"]}}
{"id": "train_07427", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "documentation"]}}
{"id": "train_07428", "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\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tooling", "tests_are_truth"]}}
{"id": "train_07429", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_07430", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_07431", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tooling"]}}
{"id": "train_07432", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "governance", "reproducibility"]}}
{"id": "train_07433", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_07434", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "repo_scale_reasoning", "tooling"]}}
{"id": "train_07435", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "ci_integration"]}}
{"id": "train_07436", "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\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "ci_integration"]}}
{"id": "train_07437", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_07438", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_07439", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "evaluation_metrics", "reproducibility"]}}
{"id": "train_07440", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "reproducibility", "evaluation_metrics"]}}
{"id": "train_07441", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tooling", "ci_integration"]}}
{"id": "train_07442", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "documentation", "tooling"]}}
{"id": "train_07443", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration"]}}
{"id": "train_07444", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "security_gates", "tooling"]}}
{"id": "train_07445", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tooling"]}}
{"id": "train_07446", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_07447", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "tooling", "tests_are_truth"]}}
{"id": "train_07448", "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\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_07449", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "evaluation_metrics", "tooling"]}}
{"id": "train_07450", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "repo_scale_reasoning", "tooling"]}}
{"id": "train_07451", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_07452", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "evaluation_metrics", "documentation"]}}
{"id": "train_07453", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "security_gates", "documentation"]}}
{"id": "train_07454", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "documentation", "repo_scale_reasoning"]}}
{"id": "train_07455", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "ci_integration", "documentation"]}}
{"id": "train_07456", "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\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "documentation", "evaluation_metrics"]}}
{"id": "train_07457", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "ci_integration", "documentation"]}}
{"id": "train_07458", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "reproducibility", "tests_are_truth"]}}
{"id": "train_07459", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_07460", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tests_are_truth", "documentation"]}}
{"id": "train_07461", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "governance", "documentation"]}}
{"id": "train_07462", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tooling", "security_gates"]}}
{"id": "train_07463", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling"]}}
{"id": "train_07464", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tooling", "reproducibility"]}}
{"id": "train_07465", "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\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_07466", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_07467", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_07468", "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\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates"]}}
{"id": "train_07469", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tooling", "evaluation_metrics"]}}
{"id": "train_07470", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "documentation"]}}
{"id": "train_07471", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "governance", "documentation"]}}
{"id": "train_07472", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "evaluation_metrics", "reproducibility"]}}
{"id": "train_07473", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "governance"]}}
{"id": "train_07474", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "reproducibility", "documentation"]}}
{"id": "train_07475", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_07476", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation"]}}
{"id": "train_07477", "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\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "documentation", "tooling"]}}
{"id": "train_07478", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "governance", "ci_integration"]}}
{"id": "train_07479", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "security_gates"]}}
{"id": "train_07480", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "reproducibility", "ci_integration"]}}
{"id": "train_07481", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "ci_integration", "governance"]}}
{"id": "train_07482", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tooling", "repo_scale_reasoning"]}}
{"id": "train_07483", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "governance"]}}
{"id": "train_07484", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "evaluation_metrics", "documentation"]}}
{"id": "train_07485", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "governance", "tooling"]}}
{"id": "train_07486", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "documentation", "governance"]}}
{"id": "train_07487", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "security_gates", "documentation"]}}
{"id": "train_07488", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_07489", "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\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "security_gates", "tests_are_truth"]}}
{"id": "train_07490", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "tests_are_truth", "reproducibility"]}}
{"id": "train_07491", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_07492", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_07493", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "reproducibility", "documentation"]}}
{"id": "train_07494", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07495", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "ci_integration", "governance"]}}
{"id": "train_07496", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "governance", "reproducibility"]}}
{"id": "train_07497", "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\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "tests_are_truth"]}}
{"id": "train_07498", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_07499", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tooling", "governance"]}}
{"id": "train_07500", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "reproducibility", "documentation"]}}
{"id": "train_07501", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tooling", "evaluation_metrics"]}}
{"id": "train_07502", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "tooling"]}}
{"id": "train_07503", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "documentation"]}}
{"id": "train_07504", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_07505", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "governance", "tests_are_truth"]}}
{"id": "train_07506", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "security_gates", "governance"]}}
{"id": "train_07507", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_07508", "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\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "security_gates", "evaluation_metrics"]}}
{"id": "train_07509", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_07510", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "security_gates", "tests_are_truth"]}}
{"id": "train_07511", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_07512", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_07513", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "evaluation_metrics"]}}
{"id": "train_07514", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "security_gates"]}}
{"id": "train_07515", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning"]}}
{"id": "train_07516", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "governance"]}}
{"id": "train_07517", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "tests_are_truth", "documentation"]}}
{"id": "train_07518", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "documentation", "security_gates"]}}
{"id": "train_07519", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tooling", "ci_integration"]}}
{"id": "train_07520", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility"]}}
{"id": "train_07521", "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\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tests_are_truth", "ci_integration"]}}
{"id": "train_07522", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics"]}}
{"id": "train_07523", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "security_gates", "tests_are_truth"]}}
{"id": "train_07524", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "governance", "repo_scale_reasoning"]}}
{"id": "train_07525", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_07526", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "documentation", "security_gates"]}}
{"id": "train_07527", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07528", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "governance", "ci_integration"]}}
{"id": "train_07529", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tests_are_truth", "documentation"]}}
{"id": "train_07530", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "security_gates", "reproducibility"]}}
{"id": "train_07531", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tooling"]}}
{"id": "train_07532", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_07533", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "security_gates", "tooling"]}}
{"id": "train_07534", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "documentation"]}}
{"id": "train_07535", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_07536", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth"]}}
{"id": "train_07537", "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\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "tooling", "documentation"]}}
{"id": "train_07538", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "tests_are_truth"]}}
{"id": "train_07539", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "ci_integration", "documentation"]}}
{"id": "train_07540", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tests_are_truth", "ci_integration"]}}
{"id": "train_07541", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "security_gates", "governance"]}}
{"id": "train_07542", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "ci_integration", "documentation"]}}
{"id": "train_07543", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "security_gates", "documentation"]}}
{"id": "train_07544", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_07545", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "governance", "documentation"]}}
{"id": "train_07546", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tests_are_truth", "ci_integration"]}}
{"id": "train_07547", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tooling", "security_gates"]}}
{"id": "train_07548", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_07549", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "documentation", "tests_are_truth"]}}
{"id": "train_07550", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "governance", "evaluation_metrics"]}}
{"id": "train_07551", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_07552", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tests_are_truth"]}}
{"id": "train_07553", "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\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration"]}}
{"id": "train_07554", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "tooling"]}}
{"id": "train_07555", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tests_are_truth"]}}
{"id": "train_07556", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tooling", "evaluation_metrics"]}}
{"id": "train_07557", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_07558", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "reproducibility"]}}
{"id": "train_07559", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07560", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning"]}}
{"id": "train_07561", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tooling", "documentation"]}}
{"id": "train_07562", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "repo_scale_reasoning", "documentation"]}}
{"id": "train_07563", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling"]}}
{"id": "train_07564", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "documentation", "tooling"]}}
{"id": "train_07565", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tooling"]}}
{"id": "train_07566", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_07567", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "reproducibility"]}}
{"id": "train_07568", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "tooling", "reproducibility"]}}
{"id": "train_07569", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "reproducibility", "tooling"]}}
{"id": "train_07570", "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\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_07571", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "ci_integration", "documentation"]}}
{"id": "train_07572", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_07573", "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\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "governance", "documentation"]}}
{"id": "train_07574", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "governance", "tests_are_truth"]}}
{"id": "train_07575", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tests_are_truth", "governance"]}}
{"id": "train_07576", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_07577", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "security_gates", "governance"]}}
{"id": "train_07578", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_07579", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "documentation", "governance"]}}
{"id": "train_07580", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates"]}}
{"id": "train_07581", "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\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07582", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "tests_are_truth", "security_gates"]}}
{"id": "train_07583", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tooling", "documentation"]}}
{"id": "train_07584", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "evaluation_metrics"]}}
{"id": "train_07585", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_07586", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "ci_integration", "evaluation_metrics"]}}
{"id": "train_07587", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_07588", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth"]}}
{"id": "train_07589", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "governance", "evaluation_metrics"]}}
{"id": "train_07590", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "governance", "security_gates"]}}
{"id": "train_07591", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "ci_integration"]}}
{"id": "train_07592", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_07593", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "governance", "tooling"]}}
{"id": "train_07594", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth"]}}
{"id": "train_07595", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation"]}}
{"id": "train_07596", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "evaluation_metrics"]}}
{"id": "train_07597", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_07598", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "evaluation_metrics", "documentation"]}}
{"id": "train_07599", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tooling"]}}
{"id": "train_07600", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_07601", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "reproducibility", "tests_are_truth"]}}
{"id": "train_07602", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_07603", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "ci_integration", "governance"]}}
{"id": "train_07604", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "reproducibility", "tooling"]}}
{"id": "train_07605", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "documentation", "ci_integration"]}}
{"id": "train_07606", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_07607", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tooling", "repo_scale_reasoning"]}}
{"id": "train_07608", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "evaluation_metrics", "governance"]}}
{"id": "train_07609", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tests_are_truth", "governance"]}}
{"id": "train_07610", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "governance"]}}
{"id": "train_07611", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tooling", "tests_are_truth"]}}
{"id": "train_07612", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_07613", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_07614", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tooling"]}}
{"id": "train_07615", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tooling"]}}
{"id": "train_07616", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tooling", "governance"]}}
{"id": "train_07617", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance"]}}
{"id": "train_07618", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "security_gates", "governance"]}}
{"id": "train_07619", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "governance", "tests_are_truth"]}}
{"id": "train_07620", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "documentation", "ci_integration"]}}
{"id": "train_07621", "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\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "ci_integration"]}}
{"id": "train_07622", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "governance", "security_gates"]}}
{"id": "train_07623", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "evaluation_metrics", "documentation"]}}
{"id": "train_07624", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "reproducibility", "ci_integration"]}}
{"id": "train_07625", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "governance", "tests_are_truth"]}}
{"id": "train_07626", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tooling", "documentation"]}}
{"id": "train_07627", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tests_are_truth", "security_gates"]}}
{"id": "train_07628", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_07629", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_07630", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_07631", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "ci_integration", "reproducibility"]}}
{"id": "train_07632", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "ci_integration"]}}
{"id": "train_07633", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "ci_integration"]}}
{"id": "train_07634", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_07635", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_07636", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_07637", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_07638", "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\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tooling", "ci_integration"]}}
{"id": "train_07639", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_07640", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "tests_are_truth"]}}
{"id": "train_07641", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_07642", "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\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_07643", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "governance"]}}
{"id": "train_07644", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "documentation", "evaluation_metrics"]}}
{"id": "train_07645", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "tests_are_truth"]}}
{"id": "train_07646", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "tooling", "reproducibility"]}}
{"id": "train_07647", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "documentation", "repo_scale_reasoning"]}}
{"id": "train_07648", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "ci_integration", "tooling"]}}
{"id": "train_07649", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "ci_integration", "reproducibility"]}}
{"id": "train_07650", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "security_gates", "governance"]}}
{"id": "train_07651", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "ci_integration", "tests_are_truth"]}}
{"id": "train_07652", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_07653", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "ci_integration", "tooling"]}}
{"id": "train_07654", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "ci_integration", "reproducibility"]}}
{"id": "train_07655", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth"]}}
{"id": "train_07656", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics"]}}
{"id": "train_07657", "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\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "documentation", "governance"]}}
{"id": "train_07658", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07659", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tooling", "evaluation_metrics"]}}
{"id": "train_07660", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration"]}}
{"id": "train_07661", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "documentation", "security_gates"]}}
{"id": "train_07662", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "security_gates", "tooling"]}}
{"id": "train_07663", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_07664", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_07665", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "security_gates", "documentation"]}}
{"id": "train_07666", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tooling", "governance"]}}
{"id": "train_07667", "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\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_07668", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_07669", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tooling"]}}
{"id": "train_07670", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth"]}}
{"id": "train_07671", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "evaluation_metrics", "governance"]}}
{"id": "train_07672", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "governance", "documentation"]}}
{"id": "train_07673", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tests_are_truth", "tooling"]}}
{"id": "train_07674", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "documentation"]}}
{"id": "train_07675", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "ci_integration", "security_gates"]}}
{"id": "train_07676", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tooling", "ci_integration"]}}
{"id": "train_07677", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_07678", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "tooling", "ci_integration"]}}
{"id": "train_07679", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_07680", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "repo_scale_reasoning", "governance"]}}
{"id": "train_07681", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "documentation", "governance"]}}
{"id": "train_07682", "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\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_07683", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_07684", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07685", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_07686", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_07687", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "governance", "reproducibility"]}}
{"id": "train_07688", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates"]}}
{"id": "train_07689", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "evaluation_metrics", "documentation"]}}
{"id": "train_07690", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "governance", "security_gates"]}}
{"id": "train_07691", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_07692", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tooling", "reproducibility"]}}
{"id": "train_07693", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "evaluation_metrics", "tooling"]}}
{"id": "train_07694", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "ci_integration"]}}
{"id": "train_07695", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_07696", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_07697", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "evaluation_metrics", "ci_integration"]}}
{"id": "train_07698", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_07699", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_07700", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_07701", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "reproducibility", "tests_are_truth"]}}
{"id": "train_07702", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "evaluation_metrics"]}}
{"id": "train_07703", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "reproducibility", "security_gates"]}}
{"id": "train_07704", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "ci_integration", "tooling"]}}
{"id": "train_07705", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_07706", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_07707", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "governance", "tooling"]}}
{"id": "train_07708", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07709", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "documentation", "tooling"]}}
{"id": "train_07710", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "ci_integration", "reproducibility"]}}
{"id": "train_07711", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "governance", "tooling"]}}
{"id": "train_07712", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_07713", "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\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_07714", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tests_are_truth", "tooling"]}}
{"id": "train_07715", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning"]}}
{"id": "train_07716", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "evaluation_metrics", "security_gates"]}}
{"id": "train_07717", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07718", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "governance", "repo_scale_reasoning"]}}
{"id": "train_07719", "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\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tests_are_truth", "reproducibility"]}}
{"id": "train_07720", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_07721", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "security_gates", "tooling"]}}
{"id": "train_07722", "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\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07723", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_07724", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "evaluation_metrics", "tooling"]}}
{"id": "train_07725", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "ci_integration", "tests_are_truth"]}}
{"id": "train_07726", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "ci_integration"]}}
{"id": "train_07727", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates"]}}
{"id": "train_07728", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "governance", "documentation"]}}
{"id": "train_07729", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tooling"]}}
{"id": "train_07730", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_07731", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "reproducibility", "documentation"]}}
{"id": "train_07732", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_07733", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_07734", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_07735", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "evaluation_metrics", "reproducibility"]}}
{"id": "train_07736", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "reproducibility", "tests_are_truth"]}}
{"id": "train_07737", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "ci_integration", "evaluation_metrics"]}}
{"id": "train_07738", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "governance"]}}
{"id": "train_07739", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_07740", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "governance", "evaluation_metrics"]}}
{"id": "train_07741", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_07742", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "documentation", "reproducibility"]}}
{"id": "train_07743", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_07744", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_07745", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_07746", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "governance"]}}
{"id": "train_07747", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "documentation", "reproducibility"]}}
{"id": "train_07748", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_07749", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "documentation"]}}
{"id": "train_07750", "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\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tooling", "security_gates"]}}
{"id": "train_07751", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling"]}}
{"id": "train_07752", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_07753", "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\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "ci_integration", "governance"]}}
{"id": "train_07754", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "ci_integration", "tests_are_truth"]}}
{"id": "train_07755", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_07756", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "documentation", "repo_scale_reasoning"]}}
{"id": "train_07757", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "governance"]}}
{"id": "train_07758", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics"]}}
{"id": "train_07759", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "documentation", "tooling"]}}
{"id": "train_07760", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "tooling", "repo_scale_reasoning"]}}
{"id": "train_07761", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tooling", "security_gates"]}}
{"id": "train_07762", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_07763", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling"]}}
{"id": "train_07764", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_07765", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_07766", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "evaluation_metrics", "ci_integration"]}}
{"id": "train_07767", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "evaluation_metrics", "governance"]}}
{"id": "train_07768", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_07769", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "governance"]}}
{"id": "train_07770", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_07771", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_07772", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_07773", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_07774", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling"]}}
{"id": "train_07775", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tests_are_truth", "tooling"]}}
{"id": "train_07776", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_07777", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07778", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "documentation", "ci_integration"]}}
{"id": "train_07779", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tooling", "governance"]}}
{"id": "train_07780", "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\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "documentation", "evaluation_metrics"]}}
{"id": "train_07781", "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\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_07782", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_07783", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "repo_scale_reasoning", "documentation"]}}
{"id": "train_07784", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_07785", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "documentation", "reproducibility"]}}
{"id": "train_07786", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_07787", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_07788", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "governance", "reproducibility"]}}
{"id": "train_07789", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "tooling"]}}
{"id": "train_07790", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tooling", "ci_integration"]}}
{"id": "train_07791", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "governance"]}}
{"id": "train_07792", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tests_are_truth", "governance"]}}
{"id": "train_07793", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "security_gates"]}}
{"id": "train_07794", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tooling", "governance"]}}
{"id": "train_07795", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "documentation", "evaluation_metrics"]}}
{"id": "train_07796", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_07797", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "ci_integration", "documentation"]}}
{"id": "train_07798", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility"]}}
{"id": "train_07799", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "documentation", "reproducibility"]}}
{"id": "train_07800", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tooling"]}}
{"id": "train_07801", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_07802", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics"]}}
{"id": "train_07803", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_07804", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration"]}}
{"id": "train_07805", "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\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_07806", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance"]}}
{"id": "train_07807", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "ci_integration", "governance"]}}
{"id": "train_07808", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tooling", "repo_scale_reasoning"]}}
{"id": "train_07809", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "ci_integration", "governance"]}}
{"id": "train_07810", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tooling", "documentation"]}}
{"id": "train_07811", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_07812", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_07813", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "tests_are_truth"]}}
{"id": "train_07814", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tooling", "tests_are_truth"]}}
{"id": "train_07815", "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\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tooling", "documentation"]}}
{"id": "train_07816", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_07817", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_07818", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_07819", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "governance"]}}
{"id": "train_07820", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_07821", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "reproducibility"]}}
{"id": "train_07822", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "documentation", "tests_are_truth"]}}
{"id": "train_07823", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_07824", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_07825", "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\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "governance", "documentation"]}}
{"id": "train_07826", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_07827", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "documentation"]}}
{"id": "train_07828", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "governance", "tooling"]}}
{"id": "train_07829", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth"]}}
{"id": "train_07830", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_07831", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "governance", "tooling"]}}
{"id": "train_07832", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tooling", "repo_scale_reasoning"]}}
{"id": "train_07833", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_07834", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "security_gates", "tooling"]}}
{"id": "train_07835", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "governance", "tests_are_truth"]}}
{"id": "train_07836", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_07837", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "tooling"]}}
{"id": "train_07838", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_07839", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_07840", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "security_gates", "governance"]}}
{"id": "train_07841", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "security_gates", "documentation"]}}
{"id": "train_07842", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "reproducibility", "tests_are_truth"]}}
{"id": "train_07843", "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\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_07844", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "documentation"]}}
{"id": "train_07845", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_07846", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_07847", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tooling", "ci_integration"]}}
{"id": "train_07848", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "ci_integration"]}}
{"id": "train_07849", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_07850", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics"]}}
{"id": "train_07851", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "tests_are_truth", "documentation"]}}
{"id": "train_07852", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "governance", "tooling"]}}
{"id": "train_07853", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "evaluation_metrics"]}}
{"id": "train_07854", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "ci_integration", "tooling"]}}
{"id": "train_07855", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "security_gates"]}}
{"id": "train_07856", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_07857", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "security_gates", "tests_are_truth"]}}
{"id": "train_07858", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "documentation"]}}
{"id": "train_07859", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "reproducibility", "tooling"]}}
{"id": "train_07860", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "ci_integration", "documentation"]}}
{"id": "train_07861", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_07862", "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\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "evaluation_metrics"]}}
{"id": "train_07863", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "documentation", "governance"]}}
{"id": "train_07864", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_07865", "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\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "documentation", "evaluation_metrics"]}}
{"id": "train_07866", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "tests_are_truth"]}}
{"id": "train_07867", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tests_are_truth", "security_gates"]}}
{"id": "train_07868", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility"]}}
{"id": "train_07869", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_07870", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "reproducibility", "tooling"]}}
{"id": "train_07871", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_07872", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "documentation", "governance"]}}
{"id": "train_07873", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tests_are_truth", "tooling"]}}
{"id": "train_07874", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_07875", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "security_gates"]}}
{"id": "train_07876", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "governance", "reproducibility"]}}
{"id": "train_07877", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "security_gates", "governance"]}}
{"id": "train_07878", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics"]}}
{"id": "train_07879", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "security_gates"]}}
{"id": "train_07880", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "tooling"]}}
{"id": "train_07881", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tests_are_truth", "tooling"]}}
{"id": "train_07882", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_07883", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics"]}}
{"id": "train_07884", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "documentation", "ci_integration"]}}
{"id": "train_07885", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tooling", "reproducibility"]}}
{"id": "train_07886", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "ci_integration"]}}
{"id": "train_07887", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "security_gates", "tooling"]}}
{"id": "train_07888", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "ci_integration", "governance"]}}
{"id": "train_07889", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling"]}}
{"id": "train_07890", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "reproducibility", "documentation"]}}
{"id": "train_07891", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "documentation", "repo_scale_reasoning"]}}
{"id": "train_07892", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "documentation"]}}
{"id": "train_07893", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "ci_integration", "governance"]}}
{"id": "train_07894", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tooling", "evaluation_metrics"]}}
{"id": "train_07895", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "evaluation_metrics", "security_gates"]}}
{"id": "train_07896", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "tests_are_truth", "governance"]}}
{"id": "train_07897", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tests_are_truth", "governance"]}}
{"id": "train_07898", "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\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "governance"]}}
{"id": "train_07899", "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\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "evaluation_metrics", "governance"]}}
{"id": "train_07900", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tests_are_truth", "tooling"]}}
{"id": "train_07901", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "governance", "tooling"]}}
{"id": "train_07902", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "tests_are_truth", "security_gates"]}}
{"id": "train_07903", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07904", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tooling", "reproducibility"]}}
{"id": "train_07905", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tooling"]}}
{"id": "train_07906", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_07907", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tooling", "evaluation_metrics"]}}
{"id": "train_07908", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility"]}}
{"id": "train_07909", "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\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "security_gates"]}}
{"id": "train_07910", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07911", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "reproducibility"]}}
{"id": "train_07912", "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\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "ci_integration", "reproducibility"]}}
{"id": "train_07913", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07914", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_07915", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_07916", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "tests_are_truth", "governance"]}}
{"id": "train_07917", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "security_gates", "ci_integration"]}}
{"id": "train_07918", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "reproducibility"]}}
{"id": "train_07919", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_07920", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "governance", "tests_are_truth"]}}
{"id": "train_07921", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "repo_scale_reasoning"]}}
{"id": "train_07922", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_07923", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_07924", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_07925", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "ci_integration", "governance"]}}
{"id": "train_07926", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_07927", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "reproducibility", "evaluation_metrics"]}}
{"id": "train_07928", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "reproducibility"]}}
{"id": "train_07929", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_07930", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_07931", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_07932", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "security_gates", "tooling"]}}
{"id": "train_07933", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "reproducibility"]}}
{"id": "train_07934", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "evaluation_metrics", "ci_integration"]}}
{"id": "train_07935", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics"]}}
{"id": "train_07936", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation"]}}
{"id": "train_07937", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "documentation", "security_gates"]}}
{"id": "train_07938", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "tests_are_truth", "governance"]}}
{"id": "train_07939", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_07940", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_07941", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "governance", "tooling"]}}
{"id": "train_07942", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_07943", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "documentation", "governance"]}}
{"id": "train_07944", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_07945", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "reproducibility", "tooling"]}}
{"id": "train_07946", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "ci_integration", "documentation"]}}
{"id": "train_07947", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "documentation", "tests_are_truth"]}}
{"id": "train_07948", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tooling"]}}
{"id": "train_07949", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_07950", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_07951", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07952", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation"]}}
{"id": "train_07953", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "documentation", "evaluation_metrics"]}}
{"id": "train_07954", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "reproducibility", "security_gates"]}}
{"id": "train_07955", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "ci_integration", "governance"]}}
{"id": "train_07956", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "governance", "evaluation_metrics"]}}
{"id": "train_07957", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "tooling", "repo_scale_reasoning"]}}
{"id": "train_07958", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "documentation", "reproducibility"]}}
{"id": "train_07959", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "governance", "tests_are_truth"]}}
{"id": "train_07960", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "governance", "tests_are_truth"]}}
{"id": "train_07961", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_07962", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_07963", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "documentation"]}}
{"id": "train_07964", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_07965", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_07966", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_07967", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "ci_integration", "tests_are_truth"]}}
{"id": "train_07968", "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\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tests_are_truth", "security_gates"]}}
{"id": "train_07969", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tests_are_truth", "security_gates"]}}
{"id": "train_07970", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_07971", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_07972", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "tests_are_truth"]}}
{"id": "train_07973", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "security_gates", "governance"]}}
{"id": "train_07974", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_07975", "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\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_07976", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "security_gates"]}}
{"id": "train_07977", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_07978", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "tooling"]}}
{"id": "train_07979", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "ci_integration"]}}
{"id": "train_07980", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning"]}}
{"id": "train_07981", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07982", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_07983", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "reproducibility", "tooling"]}}
{"id": "train_07984", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "ci_integration", "documentation"]}}
{"id": "train_07985", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_07986", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "ci_integration", "documentation"]}}
{"id": "train_07987", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_07988", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_07989", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_07990", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "tests_are_truth"]}}
{"id": "train_07991", "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\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_07992", "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\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "governance", "reproducibility"]}}
{"id": "train_07993", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tooling", "ci_integration"]}}
{"id": "train_07994", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_07995", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "reproducibility"]}}
{"id": "train_07996", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "tooling", "reproducibility"]}}
{"id": "train_07997", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "security_gates", "governance"]}}
{"id": "train_07998", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "reproducibility", "security_gates"]}}
{"id": "train_07999", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08000", "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\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "documentation", "tooling"]}}
{"id": "train_08001", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "evaluation_metrics"]}}
{"id": "train_08002", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration"]}}
{"id": "train_08003", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "security_gates", "tooling"]}}
{"id": "train_08004", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_08005", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08006", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tooling", "tests_are_truth"]}}
{"id": "train_08007", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling"]}}
{"id": "train_08008", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "reproducibility"]}}
{"id": "train_08009", "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\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_08010", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08011", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_08012", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "evaluation_metrics"]}}
{"id": "train_08013", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_08014", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "documentation"]}}
{"id": "train_08015", "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\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08016", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_08017", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "reproducibility", "documentation"]}}
{"id": "train_08018", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "documentation", "ci_integration"]}}
{"id": "train_08019", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_08020", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_08021", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_08022", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "reproducibility", "documentation"]}}
{"id": "train_08023", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tooling", "tests_are_truth"]}}
{"id": "train_08024", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_08025", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_08026", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_08027", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "security_gates"]}}
{"id": "train_08028", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "ci_integration", "documentation"]}}
{"id": "train_08029", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "tests_are_truth", "ci_integration"]}}
{"id": "train_08030", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "ci_integration", "tooling"]}}
{"id": "train_08031", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "tooling"]}}
{"id": "train_08032", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "reproducibility", "governance"]}}
{"id": "train_08033", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "ci_integration", "tooling"]}}
{"id": "train_08034", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tooling", "security_gates"]}}
{"id": "train_08035", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08036", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_08037", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08038", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "ci_integration", "security_gates"]}}
{"id": "train_08039", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "governance", "documentation"]}}
{"id": "train_08040", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "tooling", "ci_integration"]}}
{"id": "train_08041", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "governance", "reproducibility"]}}
{"id": "train_08042", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "tooling"]}}
{"id": "train_08043", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "ci_integration", "security_gates"]}}
{"id": "train_08044", "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\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "documentation", "governance"]}}
{"id": "train_08045", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "security_gates", "reproducibility"]}}
{"id": "train_08046", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "reproducibility"]}}
{"id": "train_08047", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_08048", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08049", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_08050", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "ci_integration", "tests_are_truth"]}}
{"id": "train_08051", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08052", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "evaluation_metrics", "tooling"]}}
{"id": "train_08053", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tooling", "governance"]}}
{"id": "train_08054", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "governance"]}}
{"id": "train_08055", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "evaluation_metrics", "tooling"]}}
{"id": "train_08056", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "governance", "ci_integration"]}}
{"id": "train_08057", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tooling"]}}
{"id": "train_08058", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "documentation", "reproducibility"]}}
{"id": "train_08059", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "reproducibility", "tooling"]}}
{"id": "train_08060", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tooling", "repo_scale_reasoning"]}}
{"id": "train_08061", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "tooling", "ci_integration"]}}
{"id": "train_08062", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "documentation", "security_gates"]}}
{"id": "train_08063", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "governance", "documentation"]}}
{"id": "train_08064", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "documentation", "reproducibility"]}}
{"id": "train_08065", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_08066", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_08067", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "security_gates", "evaluation_metrics"]}}
{"id": "train_08068", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tooling", "tests_are_truth"]}}
{"id": "train_08069", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_08070", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "documentation", "governance"]}}
{"id": "train_08071", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "ci_integration"]}}
{"id": "train_08072", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "tooling", "documentation"]}}
{"id": "train_08073", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tooling", "governance"]}}
{"id": "train_08074", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "reproducibility"]}}
{"id": "train_08075", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "documentation"]}}
{"id": "train_08076", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_08077", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "tooling"]}}
{"id": "train_08078", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08079", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "security_gates", "evaluation_metrics"]}}
{"id": "train_08080", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth"]}}
{"id": "train_08081", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_08082", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "governance"]}}
{"id": "train_08083", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tests_are_truth", "tooling"]}}
{"id": "train_08084", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "governance", "ci_integration"]}}
{"id": "train_08085", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08086", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning"]}}
{"id": "train_08087", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "documentation"]}}
{"id": "train_08088", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tests_are_truth", "reproducibility"]}}
{"id": "train_08089", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_08090", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_08091", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "tooling", "repo_scale_reasoning"]}}
{"id": "train_08092", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_08093", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_08094", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_08095", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility"]}}
{"id": "train_08096", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "ci_integration"]}}
{"id": "train_08097", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "governance", "ci_integration"]}}
{"id": "train_08098", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "ci_integration", "tests_are_truth"]}}
{"id": "train_08099", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "documentation", "repo_scale_reasoning"]}}
{"id": "train_08100", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_08101", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_08102", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "documentation", "governance"]}}
{"id": "train_08103", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08104", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_08105", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_08106", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tooling", "evaluation_metrics"]}}
{"id": "train_08107", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "tooling"]}}
{"id": "train_08108", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08109", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "reproducibility", "tests_are_truth"]}}
{"id": "train_08110", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_08111", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility"]}}
{"id": "train_08112", "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\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "tooling"]}}
{"id": "train_08113", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tooling", "documentation"]}}
{"id": "train_08114", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance"]}}
{"id": "train_08115", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "documentation"]}}
{"id": "train_08116", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_08117", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration"]}}
{"id": "train_08118", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_08119", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "documentation", "reproducibility"]}}
{"id": "train_08120", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08121", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance"]}}
{"id": "train_08122", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "reproducibility", "governance"]}}
{"id": "train_08123", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "governance", "ci_integration"]}}
{"id": "train_08124", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_08125", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "documentation", "evaluation_metrics"]}}
{"id": "train_08126", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "security_gates", "documentation"]}}
{"id": "train_08127", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tests_are_truth", "ci_integration"]}}
{"id": "train_08128", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_08129", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08130", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_08131", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "security_gates", "reproducibility"]}}
{"id": "train_08132", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "ci_integration"]}}
{"id": "train_08133", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_08134", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tooling", "tests_are_truth"]}}
{"id": "train_08135", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth"]}}
{"id": "train_08136", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08137", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_08138", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08139", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "security_gates", "tooling"]}}
{"id": "train_08140", "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\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "documentation", "evaluation_metrics"]}}
{"id": "train_08141", "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\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "governance"]}}
{"id": "train_08142", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_08143", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_08144", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "documentation"]}}
{"id": "train_08145", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "tooling"]}}
{"id": "train_08146", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "documentation"]}}
{"id": "train_08147", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "documentation"]}}
{"id": "train_08148", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "reproducibility", "evaluation_metrics"]}}
{"id": "train_08149", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "tooling"]}}
{"id": "train_08150", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_08151", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth"]}}
{"id": "train_08152", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "security_gates", "tooling"]}}
{"id": "train_08153", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "governance", "tooling"]}}
{"id": "train_08154", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tooling", "tests_are_truth"]}}
{"id": "train_08155", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tooling", "security_gates"]}}
{"id": "train_08156", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tooling", "tests_are_truth"]}}
{"id": "train_08157", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "tooling", "tests_are_truth"]}}
{"id": "train_08158", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling"]}}
{"id": "train_08159", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_08160", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "reproducibility", "documentation"]}}
{"id": "train_08161", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "reproducibility", "security_gates"]}}
{"id": "train_08162", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_08163", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_08164", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "documentation", "ci_integration"]}}
{"id": "train_08165", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tests_are_truth", "tooling"]}}
{"id": "train_08166", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "security_gates", "ci_integration"]}}
{"id": "train_08167", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tooling"]}}
{"id": "train_08168", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "ci_integration", "tooling"]}}
{"id": "train_08169", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_08170", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "documentation", "security_gates"]}}
{"id": "train_08171", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_08172", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "evaluation_metrics", "tooling"]}}
{"id": "train_08173", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "ci_integration", "documentation"]}}
{"id": "train_08174", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_08175", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tooling"]}}
{"id": "train_08176", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "repo_scale_reasoning", "governance"]}}
{"id": "train_08177", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "evaluation_metrics", "tooling"]}}
{"id": "train_08178", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08179", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "ci_integration", "security_gates"]}}
{"id": "train_08180", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tooling", "documentation"]}}
{"id": "train_08181", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tooling", "documentation"]}}
{"id": "train_08182", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_08183", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08184", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_08185", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08186", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "ci_integration", "security_gates"]}}
{"id": "train_08187", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_08188", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth"]}}
{"id": "train_08189", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "security_gates", "evaluation_metrics"]}}
{"id": "train_08190", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_08191", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "reproducibility"]}}
{"id": "train_08192", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08193", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "reproducibility", "ci_integration"]}}
{"id": "train_08194", "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\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_08195", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tooling", "ci_integration"]}}
{"id": "train_08196", "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\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "governance", "ci_integration"]}}
{"id": "train_08197", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_08198", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "documentation", "tooling"]}}
{"id": "train_08199", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08200", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tests_are_truth", "ci_integration"]}}
{"id": "train_08201", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "evaluation_metrics", "documentation"]}}
{"id": "train_08202", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "reproducibility", "security_gates"]}}
{"id": "train_08203", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "ci_integration"]}}
{"id": "train_08204", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "documentation", "governance"]}}
{"id": "train_08205", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "documentation"]}}
{"id": "train_08206", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "evaluation_metrics", "tooling"]}}
{"id": "train_08207", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_08208", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "security_gates", "documentation"]}}
{"id": "train_08209", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_08210", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "governance", "ci_integration"]}}
{"id": "train_08211", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_08212", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "governance", "tooling"]}}
{"id": "train_08213", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tooling", "security_gates"]}}
{"id": "train_08214", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "repo_scale_reasoning", "documentation"]}}
{"id": "train_08215", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning"]}}
{"id": "train_08216", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "documentation", "repo_scale_reasoning"]}}
{"id": "train_08217", "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\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "evaluation_metrics"]}}
{"id": "train_08218", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "governance"]}}
{"id": "train_08219", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_08220", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "ci_integration", "documentation"]}}
{"id": "train_08221", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08222", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tooling", "tests_are_truth"]}}
{"id": "train_08223", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "reproducibility"]}}
{"id": "train_08224", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "ci_integration"]}}
{"id": "train_08225", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "ci_integration", "documentation"]}}
{"id": "train_08226", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "repo_scale_reasoning", "documentation"]}}
{"id": "train_08227", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "security_gates", "tooling"]}}
{"id": "train_08228", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics"]}}
{"id": "train_08229", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tooling"]}}
{"id": "train_08230", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "ci_integration", "tests_are_truth"]}}
{"id": "train_08231", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_08232", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "security_gates", "reproducibility"]}}
{"id": "train_08233", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "governance", "tooling"]}}
{"id": "train_08234", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "tests_are_truth", "ci_integration"]}}
{"id": "train_08235", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tooling", "tests_are_truth"]}}
{"id": "train_08236", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_08237", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_08238", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance"]}}
{"id": "train_08239", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling"]}}
{"id": "train_08240", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "security_gates"]}}
{"id": "train_08241", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_08242", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08243", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_08244", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tooling", "evaluation_metrics"]}}
{"id": "train_08245", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_08246", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "repo_scale_reasoning"]}}
{"id": "train_08247", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation"]}}
{"id": "train_08248", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "governance", "ci_integration"]}}
{"id": "train_08249", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "security_gates"]}}
{"id": "train_08250", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "documentation", "evaluation_metrics"]}}
{"id": "train_08251", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "ci_integration", "reproducibility"]}}
{"id": "train_08252", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "governance"]}}
{"id": "train_08253", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "reproducibility"]}}
{"id": "train_08254", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08255", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "reproducibility", "tooling"]}}
{"id": "train_08256", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "security_gates"]}}
{"id": "train_08257", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance"]}}
{"id": "train_08258", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "documentation"]}}
{"id": "train_08259", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_08260", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "documentation", "governance"]}}
{"id": "train_08261", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance"]}}
{"id": "train_08262", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics"]}}
{"id": "train_08263", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08264", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_08265", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_08266", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08267", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_08268", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "ci_integration", "tests_are_truth"]}}
{"id": "train_08269", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "documentation", "governance"]}}
{"id": "train_08270", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "tooling", "governance"]}}
{"id": "train_08271", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_08272", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tooling", "ci_integration"]}}
{"id": "train_08273", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "ci_integration"]}}
{"id": "train_08274", "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\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_08275", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "tests_are_truth", "ci_integration"]}}
{"id": "train_08276", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08277", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "documentation", "tests_are_truth"]}}
{"id": "train_08278", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "ci_integration", "documentation"]}}
{"id": "train_08279", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "reproducibility", "evaluation_metrics"]}}
{"id": "train_08280", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tooling", "security_gates"]}}
{"id": "train_08281", "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\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "governance", "security_gates"]}}
{"id": "train_08282", "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\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tests_are_truth", "reproducibility"]}}
{"id": "train_08283", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "security_gates", "governance"]}}
{"id": "train_08284", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "security_gates", "reproducibility"]}}
{"id": "train_08285", "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\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "governance", "reproducibility"]}}
{"id": "train_08286", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_08287", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration"]}}
{"id": "train_08288", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "security_gates", "reproducibility"]}}
{"id": "train_08289", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_08290", "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\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_08291", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates"]}}
{"id": "train_08292", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_08293", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_08294", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tooling", "evaluation_metrics"]}}
{"id": "train_08295", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08296", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "governance", "documentation"]}}
{"id": "train_08297", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_08298", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration"]}}
{"id": "train_08299", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tests_are_truth", "documentation"]}}
{"id": "train_08300", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "governance", "reproducibility"]}}
{"id": "train_08301", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation"]}}
{"id": "train_08302", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_08303", "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\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "documentation", "governance"]}}
{"id": "train_08304", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "security_gates", "evaluation_metrics"]}}
{"id": "train_08305", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_08306", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_08307", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "evaluation_metrics", "security_gates"]}}
{"id": "train_08308", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "documentation"]}}
{"id": "train_08309", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "ci_integration", "documentation"]}}
{"id": "train_08310", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_08311", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "documentation", "ci_integration"]}}
{"id": "train_08312", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "security_gates", "governance"]}}
{"id": "train_08313", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_08314", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "ci_integration", "tooling"]}}
{"id": "train_08315", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "security_gates", "governance"]}}
{"id": "train_08316", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_08317", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "tooling", "ci_integration"]}}
{"id": "train_08318", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "documentation", "reproducibility"]}}
{"id": "train_08319", "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\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08320", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_08321", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "security_gates", "governance"]}}
{"id": "train_08322", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_08323", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tooling"]}}
{"id": "train_08324", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_08325", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08326", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08327", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "reproducibility", "governance"]}}
{"id": "train_08328", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tooling", "governance"]}}
{"id": "train_08329", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tooling", "ci_integration"]}}
{"id": "train_08330", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "repo_scale_reasoning", "governance"]}}
{"id": "train_08331", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "tooling"]}}
{"id": "train_08332", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tests_are_truth", "reproducibility"]}}
{"id": "train_08333", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tooling", "tests_are_truth"]}}
{"id": "train_08334", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_08335", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "ci_integration", "security_gates"]}}
{"id": "train_08336", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_08337", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_08338", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "governance"]}}
{"id": "train_08339", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tests_are_truth"]}}
{"id": "train_08340", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "documentation", "evaluation_metrics"]}}
{"id": "train_08341", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tests_are_truth"]}}
{"id": "train_08342", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "documentation", "repo_scale_reasoning"]}}
{"id": "train_08343", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_08344", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_08345", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration"]}}
{"id": "train_08346", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_08347", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_08348", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "security_gates", "governance"]}}
{"id": "train_08349", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "tooling", "repo_scale_reasoning"]}}
{"id": "train_08350", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "ci_integration", "reproducibility"]}}
{"id": "train_08351", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "governance", "reproducibility"]}}
{"id": "train_08352", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "ci_integration", "evaluation_metrics"]}}
{"id": "train_08353", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08354", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "governance", "reproducibility"]}}
{"id": "train_08355", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tooling", "governance"]}}
{"id": "train_08356", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_08357", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_08358", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tests_are_truth", "tooling"]}}
{"id": "train_08359", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tooling"]}}
{"id": "train_08360", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "ci_integration"]}}
{"id": "train_08361", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "tooling", "reproducibility"]}}
{"id": "train_08362", "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\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "documentation", "reproducibility"]}}
{"id": "train_08363", "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\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_08364", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility"]}}
{"id": "train_08365", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_08366", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "ci_integration", "tooling"]}}
{"id": "train_08367", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "evaluation_metrics", "tooling"]}}
{"id": "train_08368", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_08369", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "ci_integration", "tooling"]}}
{"id": "train_08370", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics"]}}
{"id": "train_08371", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_08372", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_08373", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth"]}}
{"id": "train_08374", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tests_are_truth", "security_gates"]}}
{"id": "train_08375", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "governance"]}}
{"id": "train_08376", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08377", "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\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics"]}}
{"id": "train_08378", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "documentation", "tests_are_truth"]}}
{"id": "train_08379", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "documentation", "tooling"]}}
{"id": "train_08380", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "reproducibility"]}}
{"id": "train_08381", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics"]}}
{"id": "train_08382", "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\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates"]}}
{"id": "train_08383", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling"]}}
{"id": "train_08384", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "reproducibility", "security_gates"]}}
{"id": "train_08385", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_08386", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "security_gates", "documentation"]}}
{"id": "train_08387", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "repo_scale_reasoning", "documentation"]}}
{"id": "train_08388", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "security_gates", "tooling"]}}
{"id": "train_08389", "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\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "tooling"]}}
{"id": "train_08390", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_08391", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_08392", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tooling", "reproducibility"]}}
{"id": "train_08393", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "governance"]}}
{"id": "train_08394", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "ci_integration"]}}
{"id": "train_08395", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "ci_integration", "security_gates"]}}
{"id": "train_08396", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "reproducibility", "governance"]}}
{"id": "train_08397", "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\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "governance"]}}
{"id": "train_08398", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tooling", "documentation"]}}
{"id": "train_08399", "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\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "documentation", "security_gates"]}}
{"id": "train_08400", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_08401", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "reproducibility", "security_gates"]}}
{"id": "train_08402", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tooling", "security_gates"]}}
{"id": "train_08403", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "security_gates"]}}
{"id": "train_08404", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "documentation", "tests_are_truth"]}}
{"id": "train_08405", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "ci_integration", "security_gates"]}}
{"id": "train_08406", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "ci_integration", "documentation"]}}
{"id": "train_08407", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "documentation"]}}
{"id": "train_08408", "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\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_08409", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "documentation", "security_gates"]}}
{"id": "train_08410", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tests_are_truth"]}}
{"id": "train_08411", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "evaluation_metrics", "documentation"]}}
{"id": "train_08412", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "reproducibility", "documentation"]}}
{"id": "train_08413", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "tooling"]}}
{"id": "train_08414", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "tests_are_truth"]}}
{"id": "train_08415", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "reproducibility", "security_gates"]}}
{"id": "train_08416", "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\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_08417", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tests_are_truth", "documentation"]}}
{"id": "train_08418", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_08419", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_08420", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics"]}}
{"id": "train_08421", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_08422", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_08423", "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\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "governance", "tests_are_truth"]}}
{"id": "train_08424", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_08425", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth"]}}
{"id": "train_08426", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tooling", "documentation"]}}
{"id": "train_08427", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "reproducibility", "security_gates"]}}
{"id": "train_08428", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "ci_integration", "tooling"]}}
{"id": "train_08429", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "tests_are_truth", "reproducibility"]}}
{"id": "train_08430", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_08431", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "documentation"]}}
{"id": "train_08432", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tooling", "security_gates"]}}
{"id": "train_08433", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "governance", "reproducibility"]}}
{"id": "train_08434", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "security_gates", "evaluation_metrics"]}}
{"id": "train_08435", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "tests_are_truth", "tooling"]}}
{"id": "train_08436", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "governance", "ci_integration"]}}
{"id": "train_08437", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "reproducibility", "documentation"]}}
{"id": "train_08438", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility"]}}
{"id": "train_08439", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_08440", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_08441", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08442", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08443", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "reproducibility", "evaluation_metrics"]}}
{"id": "train_08444", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tests_are_truth"]}}
{"id": "train_08445", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_08446", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_08447", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08448", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "governance", "ci_integration"]}}
{"id": "train_08449", "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\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_08450", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_08451", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_08452", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_08453", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_08454", "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\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "tests_are_truth"]}}
{"id": "train_08455", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "documentation", "reproducibility"]}}
{"id": "train_08456", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_08457", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "repo_scale_reasoning"]}}
{"id": "train_08458", "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\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_08459", "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\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "governance"]}}
{"id": "train_08460", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08461", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "ci_integration", "documentation"]}}
{"id": "train_08462", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_08463", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tests_are_truth"]}}
{"id": "train_08464", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "reproducibility", "documentation"]}}
{"id": "train_08465", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "evaluation_metrics"]}}
{"id": "train_08466", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_08467", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "documentation", "security_gates"]}}
{"id": "train_08468", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "ci_integration", "governance"]}}
{"id": "train_08469", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "evaluation_metrics"]}}
{"id": "train_08470", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "documentation"]}}
{"id": "train_08471", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_08472", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_08473", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tests_are_truth", "documentation"]}}
{"id": "train_08474", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "documentation", "ci_integration"]}}
{"id": "train_08475", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling"]}}
{"id": "train_08476", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_08477", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "evaluation_metrics", "tooling"]}}
{"id": "train_08478", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth"]}}
{"id": "train_08479", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "evaluation_metrics", "governance"]}}
{"id": "train_08480", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_08481", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "documentation", "tests_are_truth"]}}
{"id": "train_08482", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tooling", "documentation"]}}
{"id": "train_08483", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_08484", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_08485", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "documentation"]}}
{"id": "train_08486", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08487", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_08488", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "evaluation_metrics"]}}
{"id": "train_08489", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "reproducibility"]}}
{"id": "train_08490", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "ci_integration", "evaluation_metrics"]}}
{"id": "train_08491", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tooling"]}}
{"id": "train_08492", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "evaluation_metrics", "tooling"]}}
{"id": "train_08493", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_08494", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08495", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "evaluation_metrics"]}}
{"id": "train_08496", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08497", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "ci_integration", "evaluation_metrics"]}}
{"id": "train_08498", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "ci_integration", "reproducibility"]}}
{"id": "train_08499", "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\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tooling", "tests_are_truth"]}}
{"id": "train_08500", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tooling", "evaluation_metrics"]}}
{"id": "train_08501", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_08502", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation"]}}
{"id": "train_08503", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_08504", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_08505", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_08506", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "evaluation_metrics", "security_gates"]}}
{"id": "train_08507", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tooling", "documentation"]}}
{"id": "train_08508", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_08509", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_08510", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "security_gates", "tests_are_truth"]}}
{"id": "train_08511", "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\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tooling", "reproducibility"]}}
{"id": "train_08512", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_08513", "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\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08514", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_08515", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "ci_integration", "tests_are_truth"]}}
{"id": "train_08516", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "evaluation_metrics"]}}
{"id": "train_08517", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility"]}}
{"id": "train_08518", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates"]}}
{"id": "train_08519", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_08520", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "evaluation_metrics"]}}
{"id": "train_08521", "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\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "governance", "documentation"]}}
{"id": "train_08522", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "security_gates", "tests_are_truth"]}}
{"id": "train_08523", "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\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_08524", "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\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_08525", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "documentation", "security_gates"]}}
{"id": "train_08526", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "ci_integration", "evaluation_metrics"]}}
{"id": "train_08527", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "documentation", "repo_scale_reasoning"]}}
{"id": "train_08528", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "tests_are_truth", "governance"]}}
{"id": "train_08529", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "governance"]}}
{"id": "train_08530", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_08531", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tests_are_truth", "ci_integration"]}}
{"id": "train_08532", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "reproducibility", "tooling"]}}
{"id": "train_08533", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_08534", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tests_are_truth", "reproducibility"]}}
{"id": "train_08535", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08536", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tooling"]}}
{"id": "train_08537", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_08538", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tooling"]}}
{"id": "train_08539", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "governance", "repo_scale_reasoning"]}}
{"id": "train_08540", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "documentation", "tests_are_truth"]}}
{"id": "train_08541", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "security_gates", "evaluation_metrics"]}}
{"id": "train_08542", "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\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "governance", "security_gates"]}}
{"id": "train_08543", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "governance", "repo_scale_reasoning"]}}
{"id": "train_08544", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_08545", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_08546", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "governance", "evaluation_metrics"]}}
{"id": "train_08547", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_08548", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "ci_integration", "tooling"]}}
{"id": "train_08549", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_08550", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_08551", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tooling", "evaluation_metrics"]}}
{"id": "train_08552", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "governance", "tests_are_truth"]}}
{"id": "train_08553", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "reproducibility", "security_gates"]}}
{"id": "train_08554", "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\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_08555", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "security_gates", "governance"]}}
{"id": "train_08556", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "documentation"]}}
{"id": "train_08557", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tooling", "evaluation_metrics"]}}
{"id": "train_08558", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "ci_integration", "security_gates"]}}
{"id": "train_08559", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "ci_integration"]}}
{"id": "train_08560", "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\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "documentation", "tooling"]}}
{"id": "train_08561", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_08562", "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\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "evaluation_metrics", "ci_integration"]}}
{"id": "train_08563", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "ci_integration", "tooling"]}}
{"id": "train_08564", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "documentation", "reproducibility"]}}
{"id": "train_08565", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_08566", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "reproducibility", "tests_are_truth"]}}
{"id": "train_08567", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_08568", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "governance", "tooling"]}}
{"id": "train_08569", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "reproducibility", "evaluation_metrics"]}}
{"id": "train_08570", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling"]}}
{"id": "train_08571", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "ci_integration", "reproducibility"]}}
{"id": "train_08572", "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\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tooling"]}}
{"id": "train_08573", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_08574", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "security_gates"]}}
{"id": "train_08575", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "reproducibility", "security_gates"]}}
{"id": "train_08576", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "governance", "evaluation_metrics"]}}
{"id": "train_08577", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_08578", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08579", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "security_gates", "tests_are_truth"]}}
{"id": "train_08580", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "ci_integration", "tests_are_truth"]}}
{"id": "train_08581", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_08582", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tooling", "repo_scale_reasoning"]}}
{"id": "train_08583", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "evaluation_metrics", "governance"]}}
{"id": "train_08584", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tests_are_truth", "reproducibility"]}}
{"id": "train_08585", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "security_gates", "tests_are_truth"]}}
{"id": "train_08586", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_08587", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08588", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_08589", "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\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "reproducibility", "evaluation_metrics"]}}
{"id": "train_08590", "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\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "governance", "tooling"]}}
{"id": "train_08591", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_08592", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "reproducibility", "security_gates"]}}
{"id": "train_08593", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08594", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "reproducibility", "tooling"]}}
{"id": "train_08595", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_08596", "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\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "documentation", "tests_are_truth"]}}
{"id": "train_08597", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_08598", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "reproducibility", "governance"]}}
{"id": "train_08599", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tests_are_truth", "ci_integration"]}}
{"id": "train_08600", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "governance"]}}
{"id": "train_08601", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_08602", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08603", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "security_gates", "ci_integration"]}}
{"id": "train_08604", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_08605", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "governance"]}}
{"id": "train_08606", "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\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tests_are_truth", "reproducibility"]}}
{"id": "train_08607", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_08608", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_08609", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "reproducibility", "governance"]}}
{"id": "train_08610", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_08611", "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\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tests_are_truth"]}}
{"id": "train_08612", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "tests_are_truth", "ci_integration"]}}
{"id": "train_08613", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "governance", "tests_are_truth"]}}
{"id": "train_08614", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "repo_scale_reasoning"]}}
{"id": "train_08615", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tooling", "ci_integration"]}}
{"id": "train_08616", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_08617", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "reproducibility", "ci_integration"]}}
{"id": "train_08618", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth"]}}
{"id": "train_08619", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_08620", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tests_are_truth", "documentation"]}}
{"id": "train_08621", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth"]}}
{"id": "train_08622", "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\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_08623", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_08624", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08625", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "documentation", "repo_scale_reasoning"]}}
{"id": "train_08626", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08627", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_08628", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "security_gates", "ci_integration"]}}
{"id": "train_08629", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "ci_integration", "reproducibility"]}}
{"id": "train_08630", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_08631", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08632", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_08633", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_08634", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08635", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tooling"]}}
{"id": "train_08636", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08637", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "tooling", "reproducibility"]}}
{"id": "train_08638", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "governance", "ci_integration"]}}
{"id": "train_08639", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "documentation", "security_gates"]}}
{"id": "train_08640", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_08641", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08642", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "documentation", "security_gates"]}}
{"id": "train_08643", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "documentation", "governance"]}}
{"id": "train_08644", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "repo_scale_reasoning", "tooling"]}}
{"id": "train_08645", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08646", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance"]}}
{"id": "train_08647", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth"]}}
{"id": "train_08648", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_08649", "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\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "evaluation_metrics"]}}
{"id": "train_08650", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_08651", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tooling", "repo_scale_reasoning"]}}
{"id": "train_08652", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "tooling", "ci_integration"]}}
{"id": "train_08653", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "ci_integration"]}}
{"id": "train_08654", "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\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_08655", "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\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_08656", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "tooling", "governance"]}}
{"id": "train_08657", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_08658", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_08659", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08660", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_08661", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tests_are_truth", "ci_integration"]}}
{"id": "train_08662", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "reproducibility", "governance"]}}
{"id": "train_08663", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "evaluation_metrics"]}}
{"id": "train_08664", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "ci_integration"]}}
{"id": "train_08665", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08666", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "security_gates"]}}
{"id": "train_08667", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "governance", "security_gates"]}}
{"id": "train_08668", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "documentation", "evaluation_metrics"]}}
{"id": "train_08669", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "governance", "documentation"]}}
{"id": "train_08670", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "governance"]}}
{"id": "train_08671", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tooling", "security_gates"]}}
{"id": "train_08672", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tooling", "evaluation_metrics"]}}
{"id": "train_08673", "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\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "reproducibility", "tests_are_truth"]}}
{"id": "train_08674", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_08675", "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\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08676", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "security_gates", "documentation"]}}
{"id": "train_08677", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_08678", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "governance", "security_gates"]}}
{"id": "train_08679", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tests_are_truth", "reproducibility"]}}
{"id": "train_08680", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "ci_integration", "governance"]}}
{"id": "train_08681", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_08682", "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\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_08683", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_08684", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tooling", "evaluation_metrics"]}}
{"id": "train_08685", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "reproducibility", "documentation"]}}
{"id": "train_08686", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "governance", "evaluation_metrics"]}}
{"id": "train_08687", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "documentation"]}}
{"id": "train_08688", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tests_are_truth", "tooling"]}}
{"id": "train_08689", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "security_gates", "governance"]}}
{"id": "train_08690", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning"]}}
{"id": "train_08691", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "tooling"]}}
{"id": "train_08692", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "reproducibility", "governance"]}}
{"id": "train_08693", "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\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "reproducibility", "evaluation_metrics"]}}
{"id": "train_08694", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_08695", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "governance"]}}
{"id": "train_08696", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_08697", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tooling", "tests_are_truth"]}}
{"id": "train_08698", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "governance", "reproducibility"]}}
{"id": "train_08699", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "security_gates"]}}
{"id": "train_08700", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "documentation", "security_gates"]}}
{"id": "train_08701", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "governance", "ci_integration"]}}
{"id": "train_08702", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "documentation", "tooling"]}}
{"id": "train_08703", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_08704", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "ci_integration", "evaluation_metrics"]}}
{"id": "train_08705", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tooling", "reproducibility"]}}
{"id": "train_08706", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth"]}}
{"id": "train_08707", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["governance", "tooling", "reproducibility"]}}
{"id": "train_08708", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "documentation", "evaluation_metrics"]}}
{"id": "train_08709", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_08710", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "ci_integration", "reproducibility"]}}
{"id": "train_08711", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_08712", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "documentation", "governance"]}}
{"id": "train_08713", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_08714", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling"]}}
{"id": "train_08715", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tooling", "evaluation_metrics"]}}
{"id": "train_08716", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_08717", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "ci_integration", "evaluation_metrics"]}}
{"id": "train_08718", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tooling", "documentation"]}}
{"id": "train_08719", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_08720", "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\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "documentation", "ci_integration"]}}
{"id": "train_08721", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08722", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "security_gates", "governance"]}}
{"id": "train_08723", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08724", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_08725", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_08726", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "security_gates", "tests_are_truth"]}}
{"id": "train_08727", "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\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "governance", "repo_scale_reasoning"]}}
{"id": "train_08728", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_08729", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_08730", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "security_gates", "governance"]}}
{"id": "train_08731", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08732", "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\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "evaluation_metrics", "tooling"]}}
{"id": "train_08733", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "security_gates", "tooling"]}}
{"id": "train_08734", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "security_gates", "documentation"]}}
{"id": "train_08735", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "governance", "repo_scale_reasoning"]}}
{"id": "train_08736", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "security_gates", "evaluation_metrics"]}}
{"id": "train_08737", "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\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_08738", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_08739", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "security_gates", "evaluation_metrics"]}}
{"id": "train_08740", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08741", "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\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08742", "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\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_08743", "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\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "ci_integration"]}}
{"id": "train_08744", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08745", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "governance", "documentation"]}}
{"id": "train_08746", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "security_gates", "ci_integration"]}}
{"id": "train_08747", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_08748", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tooling", "reproducibility"]}}
{"id": "train_08749", "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\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "evaluation_metrics"]}}
{"id": "train_08750", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "security_gates"]}}
{"id": "train_08751", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08752", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_08753", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_08754", "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\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "documentation", "reproducibility"]}}
{"id": "train_08755", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "security_gates", "documentation"]}}
{"id": "train_08756", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_08757", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "reproducibility", "governance"]}}
{"id": "train_08758", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "tests_are_truth"]}}
{"id": "train_08759", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tooling"]}}
{"id": "train_08760", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_08761", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "ci_integration", "evaluation_metrics"]}}
{"id": "train_08762", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "documentation", "tooling"]}}
{"id": "train_08763", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "tooling", "governance"]}}
{"id": "train_08764", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "reproducibility", "tests_are_truth"]}}
{"id": "train_08765", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "reproducibility", "security_gates"]}}
{"id": "train_08766", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "ci_integration", "governance"]}}
{"id": "train_08767", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "ci_integration", "tooling"]}}
{"id": "train_08768", "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\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "evaluation_metrics", "documentation"]}}
{"id": "train_08769", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "repo_scale_reasoning", "tooling"]}}
{"id": "train_08770", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_08771", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_08772", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "governance"]}}
{"id": "train_08773", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_08774", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "governance", "ci_integration"]}}
{"id": "train_08775", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_08776", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tooling", "repo_scale_reasoning"]}}
{"id": "train_08777", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "documentation", "ci_integration"]}}
{"id": "train_08778", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tooling"]}}
{"id": "train_08779", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_08780", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08781", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tooling", "governance"]}}
{"id": "train_08782", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tooling", "tests_are_truth"]}}
{"id": "train_08783", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "security_gates"]}}
{"id": "train_08784", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "governance", "tests_are_truth"]}}
{"id": "train_08785", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "security_gates", "evaluation_metrics"]}}
{"id": "train_08786", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_08787", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "ci_integration", "reproducibility"]}}
{"id": "train_08788", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_08789", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08790", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08791", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "ci_integration"]}}
{"id": "train_08792", "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\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_08793", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_08794", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_08795", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "documentation", "security_gates"]}}
{"id": "train_08796", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08797", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "security_gates", "evaluation_metrics"]}}
{"id": "train_08798", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "tests_are_truth"]}}
{"id": "train_08799", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_08800", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "ci_integration", "governance"]}}
{"id": "train_08801", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_08802", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "governance", "security_gates"]}}
{"id": "train_08803", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tooling", "documentation"]}}
{"id": "train_08804", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "security_gates", "reproducibility"]}}
{"id": "train_08805", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08806", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_08807", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "documentation"]}}
{"id": "train_08808", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "governance", "evaluation_metrics"]}}
{"id": "train_08809", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "documentation", "evaluation_metrics"]}}
{"id": "train_08810", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tooling", "evaluation_metrics"]}}
{"id": "train_08811", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "tooling"]}}
{"id": "train_08812", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "governance", "documentation"]}}
{"id": "train_08813", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "governance"]}}
{"id": "train_08814", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "governance"]}}
{"id": "train_08815", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "documentation", "security_gates"]}}
{"id": "train_08816", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08817", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "tests_are_truth", "documentation"]}}
{"id": "train_08818", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_08819", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "ci_integration", "reproducibility"]}}
{"id": "train_08820", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "governance", "security_gates"]}}
{"id": "train_08821", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "documentation"]}}
{"id": "train_08822", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "repo_scale_reasoning", "tooling"]}}
{"id": "train_08823", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "evaluation_metrics"]}}
{"id": "train_08824", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_08825", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_08826", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_08827", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "ci_integration", "documentation"]}}
{"id": "train_08828", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "reproducibility"]}}
{"id": "train_08829", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_08830", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "tooling"]}}
{"id": "train_08831", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_08832", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tests_are_truth", "governance"]}}
{"id": "train_08833", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "documentation", "evaluation_metrics"]}}
{"id": "train_08834", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tooling", "repo_scale_reasoning"]}}
{"id": "train_08835", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08836", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_08837", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tooling", "governance"]}}
{"id": "train_08838", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_08839", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tooling", "evaluation_metrics"]}}
{"id": "train_08840", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08841", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "ci_integration", "evaluation_metrics"]}}
{"id": "train_08842", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tooling"]}}
{"id": "train_08843", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_08844", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tooling", "tests_are_truth"]}}
{"id": "train_08845", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "governance", "evaluation_metrics"]}}
{"id": "train_08846", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_08847", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_08848", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "ci_integration", "documentation"]}}
{"id": "train_08849", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_08850", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08851", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "documentation", "security_gates"]}}
{"id": "train_08852", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "ci_integration"]}}
{"id": "train_08853", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "ci_integration", "reproducibility"]}}
{"id": "train_08854", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_08855", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_08856", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08857", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "repo_scale_reasoning", "governance"]}}
{"id": "train_08858", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "security_gates", "evaluation_metrics"]}}
{"id": "train_08859", "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\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08860", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_08861", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "tests_are_truth"]}}
{"id": "train_08862", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "security_gates", "tests_are_truth"]}}
{"id": "train_08863", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "reproducibility", "tooling"]}}
{"id": "train_08864", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "ci_integration", "governance"]}}
{"id": "train_08865", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth"]}}
{"id": "train_08866", "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\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "governance", "evaluation_metrics"]}}
{"id": "train_08867", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "documentation"]}}
{"id": "train_08868", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "reproducibility", "security_gates"]}}
{"id": "train_08869", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "documentation", "governance"]}}
{"id": "train_08870", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "ci_integration", "reproducibility"]}}
{"id": "train_08871", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "evaluation_metrics"]}}
{"id": "train_08872", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tests_are_truth"]}}
{"id": "train_08873", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_08874", "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\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_08875", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning"]}}
{"id": "train_08876", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "ci_integration", "documentation"]}}
{"id": "train_08877", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_08878", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_08879", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "documentation", "security_gates"]}}
{"id": "train_08880", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "ci_integration", "tooling"]}}
{"id": "train_08881", "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\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "governance"]}}
{"id": "train_08882", "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\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_08883", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "ci_integration", "tooling"]}}
{"id": "train_08884", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "documentation", "governance"]}}
{"id": "train_08885", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08886", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility"]}}
{"id": "train_08887", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "tooling"]}}
{"id": "train_08888", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "reproducibility", "ci_integration"]}}
{"id": "train_08889", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "security_gates", "tooling"]}}
{"id": "train_08890", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation"]}}
{"id": "train_08891", "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\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "documentation"]}}
{"id": "train_08892", "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\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "ci_integration", "security_gates"]}}
{"id": "train_08893", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth"]}}
{"id": "train_08894", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08895", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "documentation"]}}
{"id": "train_08896", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "governance", "tooling"]}}
{"id": "train_08897", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_08898", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_08899", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_08900", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "repo_scale_reasoning", "governance"]}}
{"id": "train_08901", "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\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "evaluation_metrics", "tooling"]}}
{"id": "train_08902", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "tooling", "documentation"]}}
{"id": "train_08903", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_08904", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tooling", "ci_integration"]}}
{"id": "train_08905", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_08906", "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\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics"]}}
{"id": "train_08907", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "evaluation_metrics", "ci_integration"]}}
{"id": "train_08908", "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\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_08909", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "evaluation_metrics", "documentation"]}}
{"id": "train_08910", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_08911", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_08912", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "documentation", "governance"]}}
{"id": "train_08913", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "evaluation_metrics", "documentation"]}}
{"id": "train_08914", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "governance", "evaluation_metrics"]}}
{"id": "train_08915", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_08916", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_08917", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tooling", "documentation"]}}
{"id": "train_08918", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_08919", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_08920", "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\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "ci_integration", "governance"]}}
{"id": "train_08921", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "ci_integration", "governance"]}}
{"id": "train_08922", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_08923", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08924", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tooling", "ci_integration"]}}
{"id": "train_08925", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_08926", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_08927", "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\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "repo_scale_reasoning", "documentation"]}}
{"id": "train_08928", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "tooling", "documentation"]}}
{"id": "train_08929", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "documentation", "tests_are_truth"]}}
{"id": "train_08930", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "governance", "reproducibility"]}}
{"id": "train_08931", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_08932", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08933", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "documentation"]}}
{"id": "train_08934", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_08935", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08936", "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\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_08937", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_08938", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "security_gates", "reproducibility"]}}
{"id": "train_08939", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "governance", "security_gates"]}}
{"id": "train_08940", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08941", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_08942", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "documentation", "ci_integration"]}}
{"id": "train_08943", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tooling", "tests_are_truth"]}}
{"id": "train_08944", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_08945", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "documentation", "tests_are_truth"]}}
{"id": "train_08946", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "documentation", "repo_scale_reasoning"]}}
{"id": "train_08947", "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\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance"]}}
{"id": "train_08948", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_08949", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08950", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "evaluation_metrics", "reproducibility"]}}
{"id": "train_08951", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "evaluation_metrics"]}}
{"id": "train_08952", "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\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "reproducibility", "security_gates"]}}
{"id": "train_08953", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "documentation"]}}
{"id": "train_08954", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08955", "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\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_08956", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_08957", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "documentation", "evaluation_metrics"]}}
{"id": "train_08958", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08959", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_08960", "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\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_08961", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "documentation", "ci_integration"]}}
{"id": "train_08962", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_08963", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_08964", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_08965", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance"]}}
{"id": "train_08966", "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\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "ci_integration", "governance"]}}
{"id": "train_08967", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_08968", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "repo_scale_reasoning", "documentation"]}}
{"id": "train_08969", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth"]}}
{"id": "train_08970", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tests_are_truth", "documentation"]}}
{"id": "train_08971", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "tooling", "repo_scale_reasoning"]}}
{"id": "train_08972", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tooling", "tests_are_truth"]}}
{"id": "train_08973", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_08974", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "documentation", "tests_are_truth"]}}
{"id": "train_08975", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "documentation", "evaluation_metrics"]}}
{"id": "train_08976", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning"]}}
{"id": "train_08977", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "security_gates", "ci_integration"]}}
{"id": "train_08978", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "reproducibility", "documentation"]}}
{"id": "train_08979", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling"]}}
{"id": "train_08980", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "security_gates"]}}
{"id": "train_08981", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_08982", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tooling", "governance"]}}
{"id": "train_08983", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "documentation"]}}
{"id": "train_08984", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "ci_integration", "tooling"]}}
{"id": "train_08985", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_08986", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_08987", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "governance", "tests_are_truth"]}}
{"id": "train_08988", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_08989", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_08990", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_08991", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "tooling"]}}
{"id": "train_08992", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "ci_integration"]}}
{"id": "train_08993", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "tooling", "security_gates"]}}
{"id": "train_08994", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_08995", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "governance", "documentation"]}}
{"id": "train_08996", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling"]}}
{"id": "train_08997", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "security_gates", "governance"]}}
{"id": "train_08998", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility"]}}
{"id": "train_08999", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_09000", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tests_are_truth", "governance"]}}
{"id": "train_09001", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "tests_are_truth"]}}
{"id": "train_09002", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "documentation"]}}
{"id": "train_09003", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_09004", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "reproducibility", "tooling"]}}
{"id": "train_09005", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "governance", "tests_are_truth"]}}
{"id": "train_09006", "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\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "evaluation_metrics", "tooling"]}}
{"id": "train_09007", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "security_gates", "governance"]}}
{"id": "train_09008", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_09009", "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\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_09010", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tooling", "security_gates"]}}
{"id": "train_09011", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "ci_integration", "documentation"]}}
{"id": "train_09012", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_09013", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "security_gates", "reproducibility"]}}
{"id": "train_09014", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_09015", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "reproducibility", "documentation"]}}
{"id": "train_09016", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "tests_are_truth"]}}
{"id": "train_09017", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "governance", "tooling"]}}
{"id": "train_09018", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "governance", "reproducibility"]}}
{"id": "train_09019", "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\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "security_gates"]}}
{"id": "train_09020", "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\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "governance", "tooling"]}}
{"id": "train_09021", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tests_are_truth", "reproducibility"]}}
{"id": "train_09022", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics"]}}
{"id": "train_09023", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "repo_scale_reasoning", "documentation"]}}
{"id": "train_09024", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_09025", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_09026", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "ci_integration", "evaluation_metrics"]}}
{"id": "train_09027", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_09028", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tooling"]}}
{"id": "train_09029", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration"]}}
{"id": "train_09030", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_09031", "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\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "security_gates"]}}
{"id": "train_09032", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_09033", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "documentation", "tests_are_truth"]}}
{"id": "train_09034", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "documentation"]}}
{"id": "train_09035", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "repo_scale_reasoning", "documentation"]}}
{"id": "train_09036", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "governance"]}}
{"id": "train_09037", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "governance", "security_gates"]}}
{"id": "train_09038", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "governance", "evaluation_metrics"]}}
{"id": "train_09039", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_09040", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_09041", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_09042", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tooling", "ci_integration"]}}
{"id": "train_09043", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_09044", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_09045", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling"]}}
{"id": "train_09046", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "tests_are_truth", "ci_integration"]}}
{"id": "train_09047", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling"]}}
{"id": "train_09048", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "tooling", "tests_are_truth"]}}
{"id": "train_09049", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_09050", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_09051", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_09052", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "documentation", "reproducibility"]}}
{"id": "train_09053", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "documentation"]}}
{"id": "train_09054", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "security_gates", "governance"]}}
{"id": "train_09055", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "documentation"]}}
{"id": "train_09056", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_09057", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "reproducibility", "ci_integration"]}}
{"id": "train_09058", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_09059", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_09060", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_09061", "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\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tooling", "governance"]}}
{"id": "train_09062", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_09063", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "tooling", "evaluation_metrics"]}}
{"id": "train_09064", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "documentation"]}}
{"id": "train_09065", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "security_gates", "evaluation_metrics"]}}
{"id": "train_09066", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_09067", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_09068", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "tooling", "ci_integration"]}}
{"id": "train_09069", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tooling", "evaluation_metrics"]}}
{"id": "train_09070", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "security_gates", "documentation"]}}
{"id": "train_09071", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_09072", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "tests_are_truth"]}}
{"id": "train_09073", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "reproducibility"]}}
{"id": "train_09074", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "evaluation_metrics", "reproducibility"]}}
{"id": "train_09075", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "ci_integration", "tooling"]}}
{"id": "train_09076", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_09077", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tooling"]}}
{"id": "train_09078", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "reproducibility"]}}
{"id": "train_09079", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_09080", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_09081", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "evaluation_metrics", "governance"]}}
{"id": "train_09082", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_09083", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "tests_are_truth"]}}
{"id": "train_09084", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "governance", "security_gates"]}}
{"id": "train_09085", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_09086", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_09087", "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\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "tests_are_truth"]}}
{"id": "train_09088", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_09089", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "ci_integration", "tests_are_truth"]}}
{"id": "train_09090", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "ci_integration", "documentation"]}}
{"id": "train_09091", "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\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "evaluation_metrics"]}}
{"id": "train_09092", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "ci_integration", "reproducibility"]}}
{"id": "train_09093", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_09094", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "documentation", "evaluation_metrics"]}}
{"id": "train_09095", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_09096", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "governance", "reproducibility"]}}
{"id": "train_09097", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_09098", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_09099", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tests_are_truth", "tooling"]}}
{"id": "train_09100", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "security_gates", "governance"]}}
{"id": "train_09101", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tooling"]}}
{"id": "train_09102", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "governance", "evaluation_metrics"]}}
{"id": "train_09103", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "reproducibility"]}}
{"id": "train_09104", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "reproducibility", "governance"]}}
{"id": "train_09105", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_09106", "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\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "documentation"]}}
{"id": "train_09107", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "documentation", "governance"]}}
{"id": "train_09108", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "reproducibility", "evaluation_metrics"]}}
{"id": "train_09109", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "ci_integration", "security_gates"]}}
{"id": "train_09110", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "tooling", "documentation"]}}
{"id": "train_09111", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "governance", "reproducibility"]}}
{"id": "train_09112", "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\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "security_gates", "evaluation_metrics"]}}
{"id": "train_09113", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_09114", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_09115", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_09116", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "documentation", "governance"]}}
{"id": "train_09117", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_09118", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_09119", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tooling", "tests_are_truth"]}}
{"id": "train_09120", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_09121", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_09122", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_09123", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_09124", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "ci_integration", "tooling"]}}
{"id": "train_09125", "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\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_09126", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_09127", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling"]}}
{"id": "train_09128", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "ci_integration", "tooling"]}}
{"id": "train_09129", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "documentation", "governance"]}}
{"id": "train_09130", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "ci_integration", "evaluation_metrics"]}}
{"id": "train_09131", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_09132", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_09133", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "reproducibility", "governance"]}}
{"id": "train_09134", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_09135", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "ci_integration", "tests_are_truth"]}}
{"id": "train_09136", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tooling"]}}
{"id": "train_09137", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tooling", "tests_are_truth"]}}
{"id": "train_09138", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_09139", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tests_are_truth", "governance"]}}
{"id": "train_09140", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_09141", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["governance", "evaluation_metrics", "tooling"]}}
{"id": "train_09142", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_09143", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "security_gates", "tooling"]}}
{"id": "train_09144", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09145", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_09146", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "security_gates"]}}
{"id": "train_09147", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tests_are_truth"]}}
{"id": "train_09148", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tooling"]}}
{"id": "train_09149", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "security_gates", "tests_are_truth"]}}
{"id": "train_09150", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_09151", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling"]}}
{"id": "train_09152", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "governance", "tooling"]}}
{"id": "train_09153", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_09154", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09155", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "governance"]}}
{"id": "train_09156", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "documentation"]}}
{"id": "train_09157", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "ci_integration", "governance"]}}
{"id": "train_09158", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_09159", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_09160", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_09161", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "governance", "evaluation_metrics"]}}
{"id": "train_09162", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "governance", "evaluation_metrics"]}}
{"id": "train_09163", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_09164", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "repo_scale_reasoning", "documentation"]}}
{"id": "train_09165", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "security_gates"]}}
{"id": "train_09166", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "governance", "tests_are_truth"]}}
{"id": "train_09167", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_09168", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_09169", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "ci_integration", "reproducibility"]}}
{"id": "train_09170", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "evaluation_metrics"]}}
{"id": "train_09171", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_09172", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "security_gates", "tests_are_truth"]}}
{"id": "train_09173", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "governance", "repo_scale_reasoning"]}}
{"id": "train_09174", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tests_are_truth", "reproducibility"]}}
{"id": "train_09175", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "security_gates", "ci_integration"]}}
{"id": "train_09176", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "reproducibility", "tests_are_truth"]}}
{"id": "train_09177", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "evaluation_metrics", "governance"]}}
{"id": "train_09178", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_09179", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "tooling", "documentation"]}}
{"id": "train_09180", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth"]}}
{"id": "train_09181", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "repo_scale_reasoning", "tooling"]}}
{"id": "train_09182", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "tests_are_truth", "tooling"]}}
{"id": "train_09183", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "tests_are_truth", "ci_integration"]}}
{"id": "train_09184", "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\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "reproducibility", "governance"]}}
{"id": "train_09185", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "documentation"]}}
{"id": "train_09186", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "documentation", "security_gates"]}}
{"id": "train_09187", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tooling", "reproducibility"]}}
{"id": "train_09188", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_09189", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_09190", "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\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "documentation", "security_gates"]}}
{"id": "train_09191", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_09192", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_09193", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "reproducibility", "governance"]}}
{"id": "train_09194", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "reproducibility"]}}
{"id": "train_09195", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "documentation", "governance"]}}
{"id": "train_09196", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_09197", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_09198", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_09199", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "security_gates"]}}
{"id": "train_09200", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_09201", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_09202", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates"]}}
{"id": "train_09203", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "governance", "documentation"]}}
{"id": "train_09204", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "ci_integration", "reproducibility"]}}
{"id": "train_09205", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "security_gates", "tests_are_truth"]}}
{"id": "train_09206", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "ci_integration", "governance"]}}
{"id": "train_09207", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_09208", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "reproducibility", "evaluation_metrics"]}}
{"id": "train_09209", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "security_gates", "tests_are_truth"]}}
{"id": "train_09210", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_09211", "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\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "documentation", "ci_integration"]}}
{"id": "train_09212", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "governance", "security_gates"]}}
{"id": "train_09213", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_09214", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09215", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "repo_scale_reasoning", "tooling"]}}
{"id": "train_09216", "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\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_09217", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_09218", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_09219", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "tests_are_truth", "security_gates"]}}
{"id": "train_09220", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "security_gates", "tests_are_truth"]}}
{"id": "train_09221", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tests_are_truth"]}}
{"id": "train_09222", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_09223", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth"]}}
{"id": "train_09224", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "security_gates"]}}
{"id": "train_09225", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "reproducibility", "governance"]}}
{"id": "train_09226", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tooling"]}}
{"id": "train_09227", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "governance", "repo_scale_reasoning"]}}
{"id": "train_09228", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_09229", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_09230", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "reproducibility", "security_gates"]}}
{"id": "train_09231", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09232", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_09233", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_09234", "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\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "evaluation_metrics", "documentation"]}}
{"id": "train_09235", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration"]}}
{"id": "train_09236", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "security_gates", "ci_integration"]}}
{"id": "train_09237", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tests_are_truth", "tooling"]}}
{"id": "train_09238", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "documentation"]}}
{"id": "train_09239", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_09240", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "ci_integration", "tests_are_truth"]}}
{"id": "train_09241", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_09242", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_09243", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "documentation", "governance"]}}
{"id": "train_09244", "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\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "documentation", "security_gates"]}}
{"id": "train_09245", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tests_are_truth"]}}
{"id": "train_09246", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_09247", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "ci_integration"]}}
{"id": "train_09248", "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\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_09249", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_09250", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_09251", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tooling", "security_gates"]}}
{"id": "train_09252", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tests_are_truth", "reproducibility"]}}
{"id": "train_09253", "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\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_09254", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_09255", "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\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_09256", "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\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_09257", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "evaluation_metrics", "governance"]}}
{"id": "train_09258", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "documentation", "governance"]}}
{"id": "train_09259", "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\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "documentation", "governance"]}}
{"id": "train_09260", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "documentation"]}}
{"id": "train_09261", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates"]}}
{"id": "train_09262", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "governance", "ci_integration"]}}
{"id": "train_09263", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_09264", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "tooling", "security_gates"]}}
{"id": "train_09265", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "reproducibility", "governance"]}}
{"id": "train_09266", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_09267", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "governance", "repo_scale_reasoning"]}}
{"id": "train_09268", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth"]}}
{"id": "train_09269", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "tooling"]}}
{"id": "train_09270", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_09271", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "security_gates", "reproducibility"]}}
{"id": "train_09272", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "repo_scale_reasoning", "governance"]}}
{"id": "train_09273", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility"]}}
{"id": "train_09274", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_09275", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_09276", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "tooling", "evaluation_metrics"]}}
{"id": "train_09277", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "reproducibility", "tooling"]}}
{"id": "train_09278", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "reproducibility", "documentation"]}}
{"id": "train_09279", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_09280", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tooling", "tests_are_truth"]}}
{"id": "train_09281", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_09282", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tooling", "security_gates"]}}
{"id": "train_09283", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "security_gates", "governance"]}}
{"id": "train_09284", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "documentation", "governance"]}}
{"id": "train_09285", "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\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_09286", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "governance", "tests_are_truth"]}}
{"id": "train_09287", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "security_gates", "evaluation_metrics"]}}
{"id": "train_09288", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "ci_integration"]}}
{"id": "train_09289", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "security_gates", "evaluation_metrics"]}}
{"id": "train_09290", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_09291", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "documentation"]}}
{"id": "train_09292", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "reproducibility"]}}
{"id": "train_09293", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_09294", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_09295", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_09296", "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\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "tooling", "tests_are_truth"]}}
{"id": "train_09297", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_09298", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "governance"]}}
{"id": "train_09299", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_09300", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tests_are_truth", "documentation"]}}
{"id": "train_09301", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "evaluation_metrics", "security_gates"]}}
{"id": "train_09302", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09303", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "evaluation_metrics", "ci_integration"]}}
{"id": "train_09304", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "tests_are_truth", "documentation"]}}
{"id": "train_09305", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_09306", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "repo_scale_reasoning", "documentation"]}}
{"id": "train_09307", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09308", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tests_are_truth", "documentation", "governance"]}}
{"id": "train_09309", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_09310", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "repo_scale_reasoning", "tooling"]}}
{"id": "train_09311", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tooling"]}}
{"id": "train_09312", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_09313", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "reproducibility", "ci_integration"]}}
{"id": "train_09314", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tooling", "security_gates"]}}
{"id": "train_09315", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "governance"]}}
{"id": "train_09316", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "tests_are_truth", "governance"]}}
{"id": "train_09317", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "documentation", "governance"]}}
{"id": "train_09318", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "reproducibility", "evaluation_metrics"]}}
{"id": "train_09319", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "ci_integration", "documentation"]}}
{"id": "train_09320", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_09321", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_09322", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_09323", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "governance", "evaluation_metrics"]}}
{"id": "train_09324", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "governance", "documentation"]}}
{"id": "train_09325", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "ci_integration", "security_gates"]}}
{"id": "train_09326", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_09327", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "documentation", "tooling"]}}
{"id": "train_09328", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "governance", "documentation"]}}
{"id": "train_09329", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_09330", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_09331", "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\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "security_gates", "documentation"]}}
{"id": "train_09332", "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\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration"]}}
{"id": "train_09333", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_09334", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics"]}}
{"id": "train_09335", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tooling", "tests_are_truth"]}}
{"id": "train_09336", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_09337", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "documentation", "reproducibility"]}}
{"id": "train_09338", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_09339", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "governance", "ci_integration"]}}
{"id": "train_09340", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "reproducibility", "evaluation_metrics"]}}
{"id": "train_09341", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_09342", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_09343", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_09344", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "tests_are_truth", "reproducibility"]}}
{"id": "train_09345", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_09346", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "ci_integration"]}}
{"id": "train_09347", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_09348", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "reproducibility", "governance"]}}
{"id": "train_09349", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_09350", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tests_are_truth", "documentation"]}}
{"id": "train_09351", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "reproducibility"]}}
{"id": "train_09352", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "security_gates", "ci_integration"]}}
{"id": "train_09353", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "governance", "tests_are_truth"]}}
{"id": "train_09354", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "ci_integration", "reproducibility"]}}
{"id": "train_09355", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tests_are_truth"]}}
{"id": "train_09356", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "security_gates", "reproducibility"]}}
{"id": "train_09357", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "evaluation_metrics"]}}
{"id": "train_09358", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_09359", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "security_gates", "governance"]}}
{"id": "train_09360", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "security_gates"]}}
{"id": "train_09361", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_09362", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "ci_integration", "tooling"]}}
{"id": "train_09363", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "documentation"]}}
{"id": "train_09364", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "ci_integration", "governance"]}}
{"id": "train_09365", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_09366", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "governance", "ci_integration"]}}
{"id": "train_09367", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "evaluation_metrics", "ci_integration"]}}
{"id": "train_09368", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tests_are_truth"]}}
{"id": "train_09369", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "documentation", "governance"]}}
{"id": "train_09370", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "documentation", "evaluation_metrics"]}}
{"id": "train_09371", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "ci_integration", "tooling"]}}
{"id": "train_09372", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "governance", "documentation"]}}
{"id": "train_09373", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "documentation", "governance"]}}
{"id": "train_09374", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tooling", "security_gates"]}}
{"id": "train_09375", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "evaluation_metrics", "documentation"]}}
{"id": "train_09376", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "documentation", "reproducibility"]}}
{"id": "train_09377", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "security_gates", "reproducibility"]}}
{"id": "train_09378", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tooling", "security_gates"]}}
{"id": "train_09379", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "tooling"]}}
{"id": "train_09380", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tooling", "evaluation_metrics"]}}
{"id": "train_09381", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "ci_integration"]}}
{"id": "train_09382", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09383", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_09384", "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\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_09385", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "tooling", "documentation"]}}
{"id": "train_09386", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_09387", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility"]}}
{"id": "train_09388", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_09389", "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\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "evaluation_metrics", "documentation"]}}
{"id": "train_09390", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "reproducibility", "evaluation_metrics"]}}
{"id": "train_09391", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09392", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "documentation", "reproducibility"]}}
{"id": "train_09393", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_09394", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation"]}}
{"id": "train_09395", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_09396", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "governance", "documentation"]}}
{"id": "train_09397", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_09398", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tests_are_truth", "reproducibility"]}}
{"id": "train_09399", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_09400", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth"]}}
{"id": "train_09401", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_09402", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_09403", "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\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_09404", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "reproducibility", "tests_are_truth"]}}
{"id": "train_09405", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "documentation", "ci_integration"]}}
{"id": "train_09406", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_09407", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_09408", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "governance"]}}
{"id": "train_09409", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_09410", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "ci_integration", "documentation"]}}
{"id": "train_09411", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_09412", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "evaluation_metrics", "governance"]}}
{"id": "train_09413", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling"]}}
{"id": "train_09414", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09415", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "governance", "security_gates"]}}
{"id": "train_09416", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "governance", "reproducibility"]}}
{"id": "train_09417", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_09418", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "documentation", "ci_integration"]}}
{"id": "train_09419", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "evaluation_metrics", "tooling"]}}
{"id": "train_09420", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration"]}}
{"id": "train_09421", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_09422", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_09423", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "security_gates", "documentation"]}}
{"id": "train_09424", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "ci_integration", "documentation"]}}
{"id": "train_09425", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "ci_integration", "tooling"]}}
{"id": "train_09426", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth"]}}
{"id": "train_09427", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "reproducibility", "security_gates"]}}
{"id": "train_09428", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth"]}}
{"id": "train_09429", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_09430", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_09431", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "tooling", "evaluation_metrics"]}}
{"id": "train_09432", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_09433", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "governance", "tests_are_truth"]}}
{"id": "train_09434", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "governance"]}}
{"id": "train_09435", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "reproducibility", "evaluation_metrics"]}}
{"id": "train_09436", "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\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "ci_integration", "reproducibility"]}}
{"id": "train_09437", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_09438", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "documentation", "security_gates"]}}
{"id": "train_09439", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_09440", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "evaluation_metrics", "documentation"]}}
{"id": "train_09441", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "reproducibility", "evaluation_metrics"]}}
{"id": "train_09442", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_09443", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "documentation"]}}
{"id": "train_09444", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "ci_integration", "reproducibility"]}}
{"id": "train_09445", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["security_gates", "ci_integration", "tests_are_truth"]}}
{"id": "train_09446", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "documentation", "ci_integration"]}}
{"id": "train_09447", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "documentation", "governance"]}}
{"id": "train_09448", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_09449", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "ci_integration", "reproducibility"]}}
{"id": "train_09450", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tooling", "ci_integration"]}}
{"id": "train_09451", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning"]}}
{"id": "train_09452", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "reproducibility", "security_gates"]}}
{"id": "train_09453", "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\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "governance"]}}
{"id": "train_09454", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "ci_integration", "tests_are_truth"]}}
{"id": "train_09455", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "governance", "evaluation_metrics"]}}
{"id": "train_09456", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_09457", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "ci_integration"]}}
{"id": "train_09458", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_09459", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_09460", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "ci_integration", "security_gates"]}}
{"id": "train_09461", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "ci_integration", "governance"]}}
{"id": "train_09462", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tests_are_truth", "documentation"]}}
{"id": "train_09463", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "ci_integration", "tests_are_truth"]}}
{"id": "train_09464", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_09465", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "evaluation_metrics", "ci_integration"]}}
{"id": "train_09466", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "documentation", "reproducibility"]}}
{"id": "train_09467", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "governance", "evaluation_metrics"]}}
{"id": "train_09468", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_09469", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_09470", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09471", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "reproducibility"]}}
{"id": "train_09472", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "governance", "ci_integration"]}}
{"id": "train_09473", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tooling", "repo_scale_reasoning"]}}
{"id": "train_09474", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "tests_are_truth", "reproducibility"]}}
{"id": "train_09475", "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\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_09476", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "repo_scale_reasoning", "ci_integration"]}}
{"id": "train_09477", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tooling"]}}
{"id": "train_09478", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_09479", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates"]}}
{"id": "train_09480", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "evaluation_metrics", "reproducibility"]}}
{"id": "train_09481", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "ci_integration", "tooling"]}}
{"id": "train_09482", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_09483", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_09484", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "tooling", "governance"]}}
{"id": "train_09485", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_09486", "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\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "documentation", "security_gates"]}}
{"id": "train_09487", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "tooling"]}}
{"id": "train_09488", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "governance"]}}
{"id": "train_09489", "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\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_09490", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "documentation"]}}
{"id": "train_09491", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tests_are_truth", "governance"]}}
{"id": "train_09492", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_09493", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "security_gates", "tests_are_truth"]}}
{"id": "train_09494", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "reproducibility", "ci_integration"]}}
{"id": "train_09495", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "tooling", "ci_integration"]}}
{"id": "train_09496", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["reproducibility", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_09497", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "documentation", "ci_integration"]}}
{"id": "train_09498", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "governance", "evaluation_metrics"]}}
{"id": "train_09499", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "security_gates", "tests_are_truth"]}}
{"id": "train_09500", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "security_gates", "evaluation_metrics"]}}
{"id": "train_09501", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_09502", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "tests_are_truth", "tooling"]}}
{"id": "train_09503", "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\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "documentation", "security_gates"]}}
{"id": "train_09504", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "security_gates"]}}
{"id": "train_09505", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "security_gates", "tooling"]}}
{"id": "train_09506", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "tooling", "tests_are_truth"]}}
{"id": "train_09507", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["tests_are_truth", "evaluation_metrics", "reproducibility"]}}
{"id": "train_09508", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tests_are_truth", "tooling"]}}
{"id": "train_09509", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "security_gates", "tooling"]}}
{"id": "train_09510", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_09511", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "ci_integration", "tests_are_truth"]}}
{"id": "train_09512", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_09513", "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\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "governance", "documentation"]}}
{"id": "train_09514", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_09515", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "governance", "tests_are_truth"]}}
{"id": "train_09516", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "documentation", "evaluation_metrics"]}}
{"id": "train_09517", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "reproducibility", "tests_are_truth"]}}
{"id": "train_09518", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tooling", "governance"]}}
{"id": "train_09519", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_09520", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "documentation", "tests_are_truth"]}}
{"id": "train_09521", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "security_gates", "reproducibility"]}}
{"id": "train_09522", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_09523", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "tooling", "security_gates"]}}
{"id": "train_09524", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "security_gates"]}}
{"id": "train_09525", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tests_are_truth", "ci_integration"]}}
{"id": "train_09526", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_09527", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "security_gates", "tooling"]}}
{"id": "train_09528", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09529", "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\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "repo_scale_reasoning", "governance"]}}
{"id": "train_09530", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "governance", "security_gates"]}}
{"id": "train_09531", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "governance", "ci_integration"]}}
{"id": "train_09532", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "reproducibility", "ci_integration"]}}
{"id": "train_09533", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "governance"]}}
{"id": "train_09534", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_09535", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "governance", "tooling"]}}
{"id": "train_09536", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "reproducibility"]}}
{"id": "train_09537", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["governance", "documentation", "security_gates"]}}
{"id": "train_09538", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "documentation", "security_gates"]}}
{"id": "train_09539", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_09540", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "evaluation_metrics", "documentation"]}}
{"id": "train_09541", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_09542", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "security_gates", "reproducibility"]}}
{"id": "train_09543", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "security_gates", "ci_integration"]}}
{"id": "train_09544", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_09545", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "reproducibility", "security_gates"]}}
{"id": "train_09546", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "tooling", "reproducibility"]}}
{"id": "train_09547", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_09548", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "documentation", "tooling"]}}
{"id": "train_09549", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "documentation", "security_gates"]}}
{"id": "train_09550", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_09551", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "ci_integration"]}}
{"id": "train_09552", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "governance", "ci_integration"]}}
{"id": "train_09553", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "reproducibility", "tooling"]}}
{"id": "train_09554", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["governance", "tooling", "documentation"]}}
{"id": "train_09555", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["reproducibility", "tooling", "ci_integration"]}}
{"id": "train_09556", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "documentation", "security_gates"]}}
{"id": "train_09557", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "ci_integration", "documentation"]}}
{"id": "train_09558", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "reproducibility", "governance"]}}
{"id": "train_09559", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_09560", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "repo_scale_reasoning", "documentation"]}}
{"id": "train_09561", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "governance", "documentation"]}}
{"id": "train_09562", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "ci_integration"]}}
{"id": "train_09563", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "tests_are_truth"]}}
{"id": "train_09564", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tooling"]}}
{"id": "train_09565", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_09566", "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\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "reproducibility", "documentation"]}}
{"id": "train_09567", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "governance"]}}
{"id": "train_09568", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_09569", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "tooling", "reproducibility"]}}
{"id": "train_09570", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "governance"]}}
{"id": "train_09571", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "evaluation_metrics"]}}
{"id": "train_09572", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "security_gates", "documentation"]}}
{"id": "train_09573", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_09574", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_09575", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "documentation"]}}
{"id": "train_09576", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "repo_scale_reasoning", "cost_latency_tradeoffs"]}}
{"id": "train_09577", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["repo_scale_reasoning", "ci_integration", "reproducibility"]}}
{"id": "train_09578", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "security_gates", "governance"]}}
{"id": "train_09579", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "security_gates", "documentation"]}}
{"id": "train_09580", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09581", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_09582", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_09583", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "governance", "cost_latency_tradeoffs"]}}
{"id": "train_09584", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "security_gates", "tooling"]}}
{"id": "train_09585", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "tooling", "evaluation_metrics"]}}
{"id": "train_09586", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_09587", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_09588", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_09589", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "ci_integration", "governance"]}}
{"id": "train_09590", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Go\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "tests_are_truth", "security_gates"]}}
{"id": "train_09591", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_09592", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "tooling", "reproducibility"]}}
{"id": "train_09593", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "governance", "tooling"]}}
{"id": "train_09594", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "security_gates", "governance"]}}
{"id": "train_09595", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "ci_integration", "repo_scale_reasoning"]}}
{"id": "train_09596", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_09597", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "ci_integration", "evaluation_metrics"]}}
{"id": "train_09598", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["governance", "documentation", "ci_integration"]}}
{"id": "train_09599", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tests_are_truth", "ci_integration"]}}
{"id": "train_09600", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "reproducibility", "ci_integration"]}}
{"id": "train_09601", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "governance", "reproducibility"]}}
{"id": "train_09602", "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\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "documentation"]}}
{"id": "train_09603", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "tests_are_truth", "security_gates"]}}
{"id": "train_09604", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "documentation", "reproducibility"]}}
{"id": "train_09605", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "evaluation_metrics", "governance"]}}
{"id": "train_09606", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "security_gates", "tests_are_truth"]}}
{"id": "train_09607", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "ci_integration", "security_gates"]}}
{"id": "train_09608", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "governance", "ci_integration"]}}
{"id": "train_09609", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "tooling", "reproducibility"]}}
{"id": "train_09610", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "security_gates", "documentation"]}}
{"id": "train_09611", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["governance", "reproducibility", "documentation"]}}
{"id": "train_09612", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_09613", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_09614", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_09615", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "reproducibility", "security_gates"]}}
{"id": "train_09616", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tooling"]}}
{"id": "train_09617", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "tests_are_truth", "ci_integration"]}}
{"id": "train_09618", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "ci_integration", "tests_are_truth"]}}
{"id": "train_09619", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09620", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_09621", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "tooling", "documentation"]}}
{"id": "train_09622", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "security_gates", "tooling"]}}
{"id": "train_09623", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_09624", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_09625", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "governance", "evaluation_metrics"]}}
{"id": "train_09626", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "governance", "tooling"]}}
{"id": "train_09627", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "reproducibility"]}}
{"id": "train_09628", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Bash\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["security_gates", "documentation", "evaluation_metrics"]}}
{"id": "train_09629", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "governance"]}}
{"id": "train_09630", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "ci_integration", "tests_are_truth"]}}
{"id": "train_09631", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "tooling", "security_gates"]}}
{"id": "train_09632", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_09633", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "governance"]}}
{"id": "train_09634", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "documentation", "ci_integration"]}}
{"id": "train_09635", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "governance"]}}
{"id": "train_09636", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "evaluation_metrics", "security_gates"]}}
{"id": "train_09637", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["tooling", "documentation", "ci_integration"]}}
{"id": "train_09638", "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\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "repo_scale_reasoning", "tests_are_truth"]}}
{"id": "train_09639", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "documentation", "evaluation_metrics"]}}
{"id": "train_09640", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_09641", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tooling", "ci_integration"]}}
{"id": "train_09642", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_09643", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "governance", "reproducibility"]}}
{"id": "train_09644", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["documentation", "tests_are_truth", "reproducibility"]}}
{"id": "train_09645", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "reproducibility", "cost_latency_tradeoffs"]}}
{"id": "train_09646", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_09647", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_09648", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "security_gates", "tooling"]}}
{"id": "train_09649", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "security_gates", "evaluation_metrics"]}}
{"id": "train_09650", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "tooling", "documentation"]}}
{"id": "train_09651", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["governance", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_09652", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_09653", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: advanced\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_09654", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "governance", "reproducibility"]}}
{"id": "train_09655", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "cost_latency_tradeoffs", "security_gates"]}}
{"id": "train_09656", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "ci_integration"]}}
{"id": "train_09657", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "repo_scale_reasoning"]}}
{"id": "train_09658", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tooling"]}}
{"id": "train_09659", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tests_are_truth", "cost_latency_tradeoffs"]}}
{"id": "train_09660", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_09661", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_09662", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "governance", "security_gates"]}}
{"id": "train_09663", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_09664", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_09665", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_09666", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "ci_integration"]}}
{"id": "train_09667", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: Java\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_09668", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "governance"]}}
{"id": "train_09669", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "tooling"]}}
{"id": "train_09670", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "security_gates", "ci_integration"]}}
{"id": "train_09671", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "tests_are_truth", "repo_scale_reasoning"]}}
{"id": "train_09672", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "tests_are_truth", "governance"]}}
{"id": "train_09673", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: Go\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["ci_integration", "tests_are_truth", "tooling"]}}
{"id": "train_09674", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["documentation", "tests_are_truth", "ci_integration"]}}
{"id": "train_09675", "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\nTarget language: C#\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "repo_scale_reasoning", "tooling"]}}
{"id": "train_09676", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "security_gates", "tests_are_truth"]}}
{"id": "train_09677", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_09678", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "documentation", "governance"]}}
{"id": "train_09679", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "documentation", "cost_latency_tradeoffs"]}}
{"id": "train_09680", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_09681", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["documentation", "tooling", "ci_integration"]}}
{"id": "train_09682", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["ci_integration", "governance", "reproducibility"]}}
{"id": "train_09683", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: foundation\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "documentation", "security_gates"]}}
{"id": "train_09684", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["evaluation_metrics", "reproducibility", "documentation"]}}
{"id": "train_09685", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Bash\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "ci_integration", "governance"]}}
{"id": "train_09686", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "C#", "developer_needs": ["tests_are_truth", "governance", "security_gates"]}}
{"id": "train_09687", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "reproducibility", "ci_integration"]}}
{"id": "train_09688", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "expert", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "documentation"]}}
{"id": "train_09689", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "documentation"]}}
{"id": "train_09690", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "documentation", "security_gates"]}}
{"id": "train_09691", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "evaluation_metrics", "repo_scale_reasoning"]}}
{"id": "train_09692", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "ci_integration", "tests_are_truth"]}}
{"id": "train_09693", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "governance", "repo_scale_reasoning"]}}
{"id": "train_09694", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "documentation", "security_gates"]}}
{"id": "train_09695", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Go", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "security_gates"]}}
{"id": "train_09696", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "tests_are_truth", "documentation"]}}
{"id": "train_09697", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_09698", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["security_gates", "tests_are_truth", "governance"]}}
{"id": "train_09699", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "reproducibility", "security_gates"]}}
{"id": "train_09700", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Governance, provenance, and licensing for code data\nDifficulty: expert\nTarget language: Bash\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Bash", "developer_needs": ["reproducibility", "tooling", "security_gates"]}}
{"id": "train_09701", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["tests_are_truth", "tooling", "reproducibility"]}}
{"id": "train_09702", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["security_gates", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_09703", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_09704", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["ci_integration", "tests_are_truth", "documentation"]}}
{"id": "train_09705", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "expert", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: expert\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "security_gates", "ci_integration"]}}
{"id": "train_09706", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "tests_are_truth", "evaluation_metrics"]}}
{"id": "train_09707", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["repo_scale_reasoning", "tooling", "documentation"]}}
{"id": "train_09708", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["documentation", "security_gates", "evaluation_metrics"]}}
{"id": "train_09709", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "advanced", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "tooling", "repo_scale_reasoning"]}}
{"id": "train_09710", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_09711", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth"]}}
{"id": "train_09712", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "agent_loop", "difficulty": "foundation", "instruction": "Task: agent_loop\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "governance", "tooling"]}}
{"id": "train_09713", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "tooling"]}}
{"id": "train_09714", "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\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["ci_integration", "reproducibility", "governance"]}}
{"id": "train_09715", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "ci_integration", "tooling"]}}
{"id": "train_09716", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "documentation"]}}
{"id": "train_09717", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "expert", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: expert\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["repo_scale_reasoning", "security_gates", "tooling"]}}
{"id": "train_09718", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "evaluation_metrics", "security_gates"]}}
{"id": "train_09719", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: foundation\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "documentation", "evaluation_metrics"]}}
{"id": "train_09720", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["cost_latency_tradeoffs", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09721", "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\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "tests_are_truth", "tooling"]}}
{"id": "train_09722", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["security_gates", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_09723", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["tooling", "repo_scale_reasoning", "governance"]}}
{"id": "train_09724", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: JavaScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["governance", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_09725", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: TypeScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "security_gates", "tooling"]}}
{"id": "train_09726", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "reproducibility", "ci_integration"]}}
{"id": "train_09727", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Go\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["tests_are_truth", "tooling", "governance"]}}
{"id": "train_09728", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["tooling", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_09729", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "advanced", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "documentation", "evaluation_metrics"]}}
{"id": "train_09730", "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\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "reproducibility", "security_gates"]}}
{"id": "train_09731", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["documentation", "governance", "security_gates"]}}
{"id": "train_09732", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: SQL\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "cost_latency_tradeoffs", "evaluation_metrics"]}}
{"id": "train_09733", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["documentation", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_09734", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: advanced\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "documentation", "reproducibility"]}}
{"id": "train_09735", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Extended context and repo-scale understanding\nDifficulty: intermediate\nTarget language: Java\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["repo_scale_reasoning", "documentation", "governance"]}}
{"id": "train_09736", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["evaluation_metrics", "governance", "tooling"]}}
{"id": "train_09737", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "data_pipeline", "difficulty": "intermediate", "instruction": "Task: data_pipeline\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Python\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "tooling", "documentation"]}}
{"id": "train_09738", "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\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "ci_integration", "governance"]}}
{"id": "train_09739", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: JavaScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "tests_are_truth", "tooling"]}}
{"id": "train_09740", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["tests_are_truth", "ci_integration", "cost_latency_tradeoffs"]}}
{"id": "train_09741", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["security_gates", "tooling", "ci_integration"]}}
{"id": "train_09742", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: TypeScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tests_are_truth"]}}
{"id": "train_09743", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "documentation", "evaluation_metrics"]}}
{"id": "train_09744", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["ci_integration", "reproducibility", "governance"]}}
{"id": "train_09745", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "eval", "difficulty": "expert", "instruction": "Task: eval\nTopic: Secure code generation and policy gates\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "security_gates"]}}
{"id": "train_09746", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "explain", "difficulty": "intermediate", "instruction": "Task: explain\nTopic: Governance, provenance, and licensing for code data\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "evaluation_metrics", "governance"]}}
{"id": "train_09747", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "expert", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: expert\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Python", "developer_needs": ["tooling", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_09748", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: Rust\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "governance", "evaluation_metrics"]}}
{"id": "train_09749", "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\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tooling", "ci_integration", "evaluation_metrics"]}}
{"id": "train_09750", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "explain", "difficulty": "foundation", "instruction": "Task: explain\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "security_gates", "repo_scale_reasoning"]}}
{"id": "train_09751", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Bash", "developer_needs": ["cost_latency_tradeoffs", "ci_integration", "reproducibility"]}}
{"id": "train_09752", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "foundation", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["cost_latency_tradeoffs", "evaluation_metrics", "governance"]}}
{"id": "train_09753", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: TypeScript\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["documentation", "tests_are_truth", "ci_integration"]}}
{"id": "train_09754", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "advanced", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: advanced\nTarget language: Rust\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "ci_integration", "tooling"]}}
{"id": "train_09755", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["evaluation_metrics", "cost_latency_tradeoffs", "reproducibility"]}}
{"id": "train_09756", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Rust\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Rust", "developer_needs": ["repo_scale_reasoning", "reproducibility", "documentation"]}}
{"id": "train_09757", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_09758", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Java", "developer_needs": ["tooling", "evaluation_metrics", "tests_are_truth"]}}
{"id": "train_09759", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "foundation", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["tooling", "evaluation_metrics", "governance"]}}
{"id": "train_09760", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "reproducibility"]}}
{"id": "train_09761", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "agent_loop", "difficulty": "advanced", "instruction": "Task: agent_loop\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["tests_are_truth", "evaluation_metrics", "governance"]}}
{"id": "train_09762", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: SWE-bench style real-repo evaluation\nDifficulty: intermediate\nTarget language: Java\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Java", "developer_needs": ["cost_latency_tradeoffs", "documentation", "tooling"]}}
{"id": "train_09763", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "review", "difficulty": "expert", "instruction": "Task: review\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: expert\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "security_gates", "governance"]}}
{"id": "train_09764", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: advanced\nTarget language: Python\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReference scaffold:\n```python\ndef agent_loop(plan, edit, test, issue, max_iters=3):\n    history = []\n    p = plan(issue)\n    for _ in range(max_iters):\n        patch = edit(issue, p)\n        ok, report = test(patch)\n        history.append({\"plan\": p, \"passed\": ok, \"report\": report[:200]})\n        if ok:\n            return patch, history\n        p = p + \" | refine from failures\"\n    return patch, history\n\n```\nOperational notes: sandbox, pinned deps, human gate.", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "tooling", "ci_integration"]}}
{"id": "train_09765", "created": "2026-01-01", "topic": "Governance, provenance, and licensing for code data", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: Governance, provenance, and licensing for code data\nDifficulty: advanced\nTarget language: C#\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "evaluation_metrics", "security_gates"]}}
{"id": "train_09766", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "reproducibility", "evaluation_metrics"]}}
{"id": "train_09767", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Secure code generation and policy gates\nDifficulty: intermediate\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "cost_latency_tradeoffs", "ci_integration"]}}
{"id": "train_09768", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Extended context and repo-scale understanding\nDifficulty: expert\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "documentation", "security_gates"]}}
{"id": "train_09769", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["reproducibility", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09770", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "data_pipeline", "difficulty": "foundation", "instruction": "Task: data_pipeline\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Java\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Java", "developer_needs": ["evaluation_metrics", "reproducibility", "tooling"]}}
{"id": "train_09771", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["governance", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09772", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Extended context and repo-scale understanding\nDifficulty: foundation\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["cost_latency_tradeoffs", "reproducibility", "repo_scale_reasoning"]}}
{"id": "train_09773", "created": "2026-01-01", "topic": "Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)", "task_type": "design", "difficulty": "intermediate", "instruction": "Task: design\nTopic: Agentic coding systems (plan\u2192edit\u2192test\u2192reflect)\nDifficulty: intermediate\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["cost_latency_tradeoffs", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_09774", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: expert\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_09775", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "agent_loop", "difficulty": "expert", "instruction": "Task: agent_loop\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "reproducibility", "documentation"]}}
{"id": "train_09776", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: intermediate\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["repo_scale_reasoning", "evaluation_metrics", "documentation"]}}
{"id": "train_09777", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "review", "difficulty": "intermediate", "instruction": "Task: review\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: TypeScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "TypeScript", "developer_needs": ["ci_integration", "governance", "security_gates"]}}
{"id": "train_09778", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: foundation\nTarget language: SQL\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["security_gates", "tooling", "tests_are_truth"]}}
{"id": "train_09779", "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\nTarget language: Java\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Java", "developer_needs": ["documentation", "governance", "repo_scale_reasoning"]}}
{"id": "train_09780", "created": "2026-01-01", "topic": "Reasoning-first coding models and tunable deliberation", "task_type": "agent_loop", "difficulty": "intermediate", "instruction": "Task: agent_loop\nTopic: Reasoning-first coding models and tunable deliberation\nDifficulty: intermediate\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "Rust", "developer_needs": ["ci_integration", "documentation", "tooling"]}}
{"id": "train_09781", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Go\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Go", "developer_needs": ["documentation", "reproducibility", "ci_integration"]}}
{"id": "train_09782", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "design", "difficulty": "foundation", "instruction": "Task: design\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: JavaScript\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["tests_are_truth", "governance", "documentation"]}}
{"id": "train_09783", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: advanced\nTarget language: Python\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Python", "developer_needs": ["ci_integration", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_09784", "created": "2026-01-01", "topic": "Mixture-of-Experts (MoE) for code", "task_type": "eval", "difficulty": "intermediate", "instruction": "Task: eval\nTopic: Mixture-of-Experts (MoE) for code\nDifficulty: intermediate\nTarget language: Rust\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Rust", "developer_needs": ["security_gates", "repo_scale_reasoning", "tooling"]}}
{"id": "train_09785", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "intermediate", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: intermediate\nTarget language: Bash\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "reproducibility", "documentation"]}}
{"id": "train_09786", "created": "2026-01-01", "topic": "Code-specialized model families and sizing tradeoffs", "task_type": "eval", "difficulty": "foundation", "instruction": "Task: eval\nTopic: Code-specialized model families and sizing tradeoffs\nDifficulty: foundation\nTarget language: Python\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "Python", "developer_needs": ["reproducibility", "cost_latency_tradeoffs", "repo_scale_reasoning"]}}
{"id": "train_09787", "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\nTarget language: JavaScript\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["evaluation_metrics", "governance", "reproducibility"]}}
{"id": "train_09788", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: expert\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "reproducibility", "ci_integration"]}}
{"id": "train_09789", "created": "2026-01-01", "topic": "Dataset curation pipelines (filter, dedupe, quality)", "task_type": "compare", "difficulty": "expert", "instruction": "Task: compare\nTopic: Dataset curation pipelines (filter, dedupe, quality)\nDifficulty: expert\nTarget language: SQL\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "SQL", "developer_needs": ["ci_integration", "evaluation_metrics", "governance"]}}
{"id": "train_09790", "created": "2026-01-01", "topic": "Tool calling, sandboxes, and CI integration", "task_type": "code", "difficulty": "intermediate", "instruction": "Task: code\nTopic: Tool calling, sandboxes, and CI integration\nDifficulty: intermediate\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "Bash", "developer_needs": ["governance", "tests_are_truth", "tooling"]}}
{"id": "train_09791", "created": "2026-01-01", "topic": "Model merging, distillation, and continued pretraining", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Model merging, distillation, and continued pretraining\nDifficulty: foundation\nTarget language: Bash\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Bash", "developer_needs": ["tooling", "tests_are_truth", "ci_integration"]}}
{"id": "train_09792", "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\nTarget language: SQL\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nEval:\n- Tasks: real issues\n- Metrics: pass@k, time-to-green\n- Gates: lint/security\n", "metadata": {"target_language": "SQL", "developer_needs": ["reproducibility", "evaluation_metrics", "cost_latency_tradeoffs"]}}
{"id": "train_09793", "created": "2026-01-01", "topic": "Extended context and repo-scale understanding", "task_type": "code", "difficulty": "advanced", "instruction": "Task: code\nTopic: Extended context and repo-scale understanding\nDifficulty: advanced\nTarget language: C#\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "C#", "developer_needs": ["governance", "repo_scale_reasoning", "evaluation_metrics"]}}
{"id": "train_09794", "created": "2026-01-01", "topic": "Multimodal dev workflows (docs, diagrams, traces)", "task_type": "review", "difficulty": "foundation", "instruction": "Task: review\nTopic: Multimodal dev workflows (docs, diagrams, traces)\nDifficulty: foundation\nTarget language: C#\nContext: Fix a failing issue with tests as the oracle and produce a safe patch.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "C#", "developer_needs": ["reproducibility", "tooling", "documentation"]}}
{"id": "train_09795", "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\nTarget language: SQL\nContext: Integrate an LLM agent into CI for a large monorepo.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nAgent loop: Plan \u2192 Edit (diff) \u2192 Test \u2192 Reflect \u2192 Human gate\n", "metadata": {"target_language": "SQL", "developer_needs": ["tests_are_truth", "documentation", "repo_scale_reasoning"]}}
{"id": "train_09796", "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\nTarget language: Go\nContext: Create an eval harness that reflects real developer workflows.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nCompare: capability, cost, latency, reliability, governance\n", "metadata": {"target_language": "Go", "developer_needs": ["evaluation_metrics", "governance", "documentation"]}}
{"id": "train_09797", "created": "2026-01-01", "topic": "Secure code generation and policy gates", "task_type": "design", "difficulty": "advanced", "instruction": "Task: design\nTopic: Secure code generation and policy gates\nDifficulty: advanced\nTarget language: JavaScript\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nDesign guidance with risks, metrics, acceptance criteria\n", "metadata": {"target_language": "JavaScript", "developer_needs": ["repo_scale_reasoning", "security_gates", "cost_latency_tradeoffs"]}}
{"id": "train_09798", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "data_pipeline", "difficulty": "advanced", "instruction": "Task: data_pipeline\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Rust\nContext: Design a data pipeline for continued pretraining with auditability.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nPipeline:\n1) Ingest\n2) Normalize\n3) Filter\n4) Dedupe\n5) Quality score\n6) Sample\n7) Audit\n", "metadata": {"target_language": "Rust", "developer_needs": ["reproducibility", "tooling", "cost_latency_tradeoffs"]}}
{"id": "train_09799", "created": "2026-01-01", "topic": "SWE-bench style real-repo evaluation", "task_type": "review", "difficulty": "advanced", "instruction": "Task: review\nTopic: SWE-bench style real-repo evaluation\nDifficulty: advanced\nTarget language: Python\nContext: Evaluate two coding models for internal rollout under strict governance.\nDeliver production-grade guidance or artifacts.", "input": "", "output": "Key facts:\n- Modern AI coding emphasizes correctness via tests, agentic loops, and real-repo evaluation.\n- Reasoning-first and MoE approaches improve capability-per-compute when paired with tools.\n\nReview: correctness, security, performance, governance\n", "metadata": {"target_language": "Python", "developer_needs": ["evaluation_metrics", "security_gates", "reproducibility"]}}