id
stringlengths
12
12
created
timestamp[s]date
2026-01-01 00:00:00
2026-01-01 00:00:00
topic
stringclasses
15 values
task_type
stringclasses
12 values
difficulty
stringclasses
2 values
instruction
stringlengths
176
230
input
stringclasses
1 value
output
stringclasses
11 values
metadata
dict
hash
stringlengths
40
40
train_015000
2026-01-01T00:00:00
Governance, provenance, and audit logging
data_pipeline
advanced
Task: data_pipeline Topic: Governance, provenance, and audit logging Difficulty: advanced Language: JavaScript Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "JavaScript", "moe_experts": [ "coding", "agent" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
6454c74b02acd946388ce583976c51a3421fdc80
train_015001
2026-01-01T00:00:00
Governance, provenance, and audit logging
patch_diff
advanced
Task: patch_diff Topic: Governance, provenance, and audit logging Difficulty: advanced Language: JavaScript Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Patch: ```diff - result = a / b + if b == 0: + raise ValueError('division by zero') + result = a / b ```
{ "language": "JavaScript", "moe_experts": [ "agent", "performance" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
77c883fa3673c073f1303ddb5dde3ed94fcfd311
train_015002
2026-01-01T00:00:00
Secure code generation, red-teaming, and policy gates
agent_loop
advanced
Task: agent_loop Topic: Secure code generation, red-teaming, and policy gates Difficulty: advanced Language: TypeScript Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "TypeScript", "moe_experts": [ "agent", "coding" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
d1030725953ee432a1643bd6f4adefbc9a594fe0
train_015003
2026-01-01T00:00:00
Dataset curation pipelines and quality attribution
agent_loop
expert
Task: agent_loop Topic: Dataset curation pipelines and quality attribution Difficulty: expert Language: Rust Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Rust", "moe_experts": [ "research", "performance" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
61c133795dbf7349f0ae6c5d57a04830636dce99
train_015004
2026-01-01T00:00:00
Self-grading and automated model critique
data_pipeline
advanced
Task: data_pipeline Topic: Self-grading and automated model critique Difficulty: advanced Language: Rust Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Rust", "moe_experts": [ "coding", "research" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
0073d9f517528c565ec725ac49f21a6e11dd687d
train_015005
2026-01-01T00:00:00
Code-specialized model families and scaling laws
failure_analysis
advanced
Task: failure_analysis Topic: Code-specialized model families and scaling laws Difficulty: advanced Language: TypeScript Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Failure: regression detected Cause: missing edge case Fix: add guard + test
{ "language": "TypeScript", "moe_experts": [ "governance", "security" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
276ed2d17cbca25ac2d126c95f5beb87adb73ba0
train_015006
2026-01-01T00:00:00
Mixture-of-Experts (MoE) routing and sparsity
eval
advanced
Task: eval Topic: Mixture-of-Experts (MoE) routing and sparsity Difficulty: advanced Language: Python Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Metrics: - pass@k - time-to-green - cost-per-fix - regression rate
{ "language": "Python", "moe_experts": [ "agent", "performance" ], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true }
420c3c60622796f5843b76b5fb18d0d0f925ced3
train_015007
2026-01-01T00:00:00
Tool calling protocols and sandboxed execution
design
expert
Task: design Topic: Tool calling protocols and sandboxed execution Difficulty: expert Language: C# Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "C#", "moe_experts": [ "data", "security" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
36d1afb3beb8d5a5f273fdb10be6d38898b1143d
train_015008
2026-01-01T00:00:00
Dataset curation pipelines and quality attribution
compare
expert
Task: compare Topic: Dataset curation pipelines and quality attribution Difficulty: expert Language: Go Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Go", "moe_experts": [ "security", "data" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
6925d6167c8f805aa13bc9ce358f4436562d42ba
train_015009
2026-01-01T00:00:00
Agentic coding systems (plan→edit→test→reflect)
code
expert
Task: code Topic: Agentic coding systems (plan→edit→test→reflect) Difficulty: expert Language: Bash Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Bash", "moe_experts": [ "evaluation", "security" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
4e5c3f4b46e0dec070e0983099c609f392f12194
train_015010
2026-01-01T00:00:00
Failure engineering and resilience testing
review
advanced
Task: review Topic: Failure engineering and resilience testing Difficulty: advanced Language: Go Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Go", "moe_experts": [ "research", "performance" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
da6371b5c5440d96a5a2c88d454f406685902d94
train_015011
2026-01-01T00:00:00
Model merging, distillation, and continual learning
data_pipeline
advanced
Task: data_pipeline Topic: Model merging, distillation, and continual learning Difficulty: advanced Language: SQL Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "SQL", "moe_experts": [ "coding", "security" ], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true }
5200801a6b22b8a4ba7813b1b65374fd229886d2
train_015012
2026-01-01T00:00:00
Failure engineering and resilience testing
self_grade
advanced
Task: self_grade Topic: Failure engineering and resilience testing Difficulty: advanced Language: Java Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. { "score": 8, "confidence": 0.8, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "Java", "moe_experts": [ "performance", "governance" ], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true }
8411fc45e9479c2751b73d8a2d2bb5d96f6b1557
train_015013
2026-01-01T00:00:00
Agentic coding systems (plan→edit→test→reflect)
agent_loop
expert
Task: agent_loop Topic: Agentic coding systems (plan→edit→test→reflect) Difficulty: expert Language: JavaScript Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "JavaScript", "moe_experts": [ "research", "coding" ], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true }
be705a76348c3461ad58c9b69183f5078c6602b4
train_015014
2026-01-01T00:00:00
Agentic coding systems (plan→edit→test→reflect)
tool_trace
advanced
Task: tool_trace Topic: Agentic coding systems (plan→edit→test→reflect) Difficulty: advanced Language: Python Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "Python", "moe_experts": [ "research", "governance" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
d82aad6c2062bef104443a454f96557d9a359f75
train_015015
2026-01-01T00:00:00
Latency, cost, throughput optimization
code
expert
Task: code Topic: Latency, cost, throughput optimization Difficulty: expert Language: C# Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "C#", "moe_experts": [ "evaluation", "coding" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
2f2650310de2378a47a360bf7b6872e756c2af9b
train_015016
2026-01-01T00:00:00
Reasoning-first coding models and tunable deliberation
compare
advanced
Task: compare Topic: Reasoning-first coding models and tunable deliberation Difficulty: advanced Language: TypeScript Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "TypeScript", "moe_experts": [ "agent", "research" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
5b3d597eb88cf9320f7de377351e0f4481f54149
train_015017
2026-01-01T00:00:00
Dataset curation pipelines and quality attribution
data_pipeline
advanced
Task: data_pipeline Topic: Dataset curation pipelines and quality attribution Difficulty: advanced Language: SQL Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "SQL", "moe_experts": [ "security", "coding" ], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true }
0a7aa09f9c75f45732f57808c5fce66c79191e7d
train_015018
2026-01-01T00:00:00
Latency, cost, throughput optimization
failure_analysis
expert
Task: failure_analysis Topic: Latency, cost, throughput optimization Difficulty: expert Language: SQL Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Failure: regression detected Cause: missing edge case Fix: add guard + test
{ "language": "SQL", "moe_experts": [ "governance", "evaluation" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
fda512b076a9832c8d5b5ba29fe3fc7f2948eb3b
train_015019
2026-01-01T00:00:00
Agentic coding systems (plan→edit→test→reflect)
data_pipeline
advanced
Task: data_pipeline Topic: Agentic coding systems (plan→edit→test→reflect) Difficulty: advanced Language: JavaScript Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "JavaScript", "moe_experts": [ "research", "evaluation" ], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true }
926ca0d356130117ba6395d8a02949220fa435e4
train_015020
2026-01-01T00:00:00
Self-grading and automated model critique
compare
expert
Task: compare Topic: Self-grading and automated model critique Difficulty: expert Language: JavaScript Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "JavaScript", "moe_experts": [ "governance", "security" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
66df9f989e1a5065439e599c7e060b000287c139
train_015021
2026-01-01T00:00:00
Self-grading and automated model critique
compare
advanced
Task: compare Topic: Self-grading and automated model critique Difficulty: advanced Language: Bash Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Bash", "moe_experts": [ "performance", "governance" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
df1fdc2b58f00d0e8b3527fd0864d244e0878370
train_015022
2026-01-01T00:00:00
Model merging, distillation, and continual learning
self_grade
advanced
Task: self_grade Topic: Model merging, distillation, and continual learning Difficulty: advanced Language: C# Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. { "score": 8, "confidence": 0.8, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "C#", "moe_experts": [ "agent", "research" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
bdcdb3ef8501ae959081c882b72c56c8b8512be8
train_015023
2026-01-01T00:00:00
Reasoning-first coding models and tunable deliberation
agent_loop
expert
Task: agent_loop Topic: Reasoning-first coding models and tunable deliberation Difficulty: expert Language: Rust Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Rust", "moe_experts": [ "coding", "data" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
6232a9557517df5d435b6eb6c06cd1093edad9cf
train_015024
2026-01-01T00:00:00
Dataset curation pipelines and quality attribution
agent_loop
expert
Task: agent_loop Topic: Dataset curation pipelines and quality attribution Difficulty: expert Language: SQL Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "SQL", "moe_experts": [ "governance", "coding" ], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true }
8f4904588fa27d8ab54b9ee15ec00549c5f59da0
train_015025
2026-01-01T00:00:00
Secure code generation, red-teaming, and policy gates
self_grade
expert
Task: self_grade Topic: Secure code generation, red-teaming, and policy gates Difficulty: expert Language: SQL Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. { "score": 10, "confidence": 1.0, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "SQL", "moe_experts": [ "data", "coding" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
49e2172b1922217a8df27774bddf94dac182121a
train_015026
2026-01-01T00:00:00
Reasoning-first coding models and tunable deliberation
eval
expert
Task: eval Topic: Reasoning-first coding models and tunable deliberation Difficulty: expert Language: SQL Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Metrics: - pass@k - time-to-green - cost-per-fix - regression rate
{ "language": "SQL", "moe_experts": [ "coding", "data" ], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true }
797a54c93bb0c4a3b7e666fcb31beebe13a9673b
train_015027
2026-01-01T00:00:00
Secure code generation, red-teaming, and policy gates
design
advanced
Task: design Topic: Secure code generation, red-teaming, and policy gates Difficulty: advanced Language: TypeScript Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "TypeScript", "moe_experts": [ "governance", "research" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
c418adbe68abc65b5170adf9bfedcbbbd23b05b4
train_015028
2026-01-01T00:00:00
Latency, cost, throughput optimization
agent_loop
advanced
Task: agent_loop Topic: Latency, cost, throughput optimization Difficulty: advanced Language: TypeScript Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "TypeScript", "moe_experts": [ "performance", "research" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
82b03c12f982747cbc1e2f74ead3fe0b9ffc4e45
train_015029
2026-01-01T00:00:00
Model merging, distillation, and continual learning
explain
advanced
Task: explain Topic: Model merging, distillation, and continual learning Difficulty: advanced Language: Bash Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Bash", "moe_experts": [ "evaluation", "data" ], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true }
c83eb90c8f4ad09cdc25bf48094b3e2ba0da0653
train_015030
2026-01-01T00:00:00
Code-specialized model families and scaling laws
eval
advanced
Task: eval Topic: Code-specialized model families and scaling laws Difficulty: advanced Language: Bash Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Metrics: - pass@k - time-to-green - cost-per-fix - regression rate
{ "language": "Bash", "moe_experts": [ "governance", "security" ], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true }
6c8a0675c04d1c9b14bdc3892b9738e88acc1491
train_015031
2026-01-01T00:00:00
Latency, cost, throughput optimization
data_pipeline
advanced
Task: data_pipeline Topic: Latency, cost, throughput optimization Difficulty: advanced Language: Python Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Python", "moe_experts": [ "evaluation", "data" ], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true }
518df62129561b5fe11513da8f6254be784d5f45
train_015032
2026-01-01T00:00:00
Governance, provenance, and audit logging
compare
expert
Task: compare Topic: Governance, provenance, and audit logging Difficulty: expert Language: Go Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Go", "moe_experts": [ "performance", "coding" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
c970982f979f3086122a23d78cd6703a7cdd71d4
train_015033
2026-01-01T00:00:00
Model merging, distillation, and continual learning
design
expert
Task: design Topic: Model merging, distillation, and continual learning Difficulty: expert Language: Python Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Python", "moe_experts": [ "agent", "governance" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
b7f133462255cb544130ef5fba2ca956c882ec6f
train_015034
2026-01-01T00:00:00
Mixture-of-Experts (MoE) routing and sparsity
self_grade
expert
Task: self_grade Topic: Mixture-of-Experts (MoE) routing and sparsity Difficulty: expert Language: SQL Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. { "score": 10, "confidence": 1.0, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "SQL", "moe_experts": [ "research", "security" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
93a4d5b9d7dd22c5224bc04797a346e14519e672
train_015035
2026-01-01T00:00:00
Failure engineering and resilience testing
compare
advanced
Task: compare Topic: Failure engineering and resilience testing Difficulty: advanced Language: Bash Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Bash", "moe_experts": [ "agent", "performance" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
437b9a0984fefcec057d05207211b268af16fb5d
train_015036
2026-01-01T00:00:00
Mixture-of-Experts (MoE) routing and sparsity
tool_trace
advanced
Task: tool_trace Topic: Mixture-of-Experts (MoE) routing and sparsity Difficulty: advanced Language: SQL Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "SQL", "moe_experts": [ "coding", "research" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
af8780f9d14a76d19a61637ff9a619b7282bc15e
train_015037
2026-01-01T00:00:00
Extended context, memory, and repo graphs
data_pipeline
advanced
Task: data_pipeline Topic: Extended context, memory, and repo graphs Difficulty: advanced Language: SQL Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "SQL", "moe_experts": [ "evaluation", "research" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
be68a1e3f4c7c35d319b0f7d96c8101b07d07d26
train_015038
2026-01-01T00:00:00
Secure code generation, red-teaming, and policy gates
agent_loop
expert
Task: agent_loop Topic: Secure code generation, red-teaming, and policy gates Difficulty: expert Language: Python Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Python", "moe_experts": [ "coding", "data" ], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true }
cdcfc11a5a454f82798c34b8f0f94db1687b51c5
train_015039
2026-01-01T00:00:00
Extended context, memory, and repo graphs
data_pipeline
advanced
Task: data_pipeline Topic: Extended context, memory, and repo graphs Difficulty: advanced Language: Bash Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Bash", "moe_experts": [ "coding", "research" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
60969c7cc54d0c505ed1ecce8e5c0b33d94f98b0
train_015040
2026-01-01T00:00:00
Tool calling protocols and sandboxed execution
tool_trace
advanced
Task: tool_trace Topic: Tool calling protocols and sandboxed execution Difficulty: advanced Language: JavaScript Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "JavaScript", "moe_experts": [ "performance", "coding" ], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true }
608055982bf400f0a2a71ebd49e8313255606acc
train_015041
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
explain
advanced
Task: explain Topic: SWE-bench real-repo evaluation and leakage control Difficulty: advanced Language: JavaScript Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "JavaScript", "moe_experts": [ "security", "performance" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
b958b9bd0efd2b33528129db7a1f9cc1640a61d6
train_015042
2026-01-01T00:00:00
Code-specialized model families and scaling laws
data_pipeline
expert
Task: data_pipeline Topic: Code-specialized model families and scaling laws Difficulty: expert Language: JavaScript Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "JavaScript", "moe_experts": [ "coding", "performance" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
577cabf3ea5d5f1d2c600387583f620129bd280b
train_015043
2026-01-01T00:00:00
Extended context, memory, and repo graphs
design
expert
Task: design Topic: Extended context, memory, and repo graphs Difficulty: expert Language: Python Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Python", "moe_experts": [ "coding", "performance" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
f707ce89dd15067ba44c455c4291ac91143fb3f4
train_015044
2026-01-01T00:00:00
Multi-agent coordination and role separation
tool_trace
expert
Task: tool_trace Topic: Multi-agent coordination and role separation Difficulty: expert Language: Java Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "Java", "moe_experts": [ "performance", "agent" ], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true }
6d98e5d299d69e639e9ca9ba0b067597be3c4037
train_015045
2026-01-01T00:00:00
Tool calling protocols and sandboxed execution
tool_trace
expert
Task: tool_trace Topic: Tool calling protocols and sandboxed execution Difficulty: expert Language: Go Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "Go", "moe_experts": [ "performance", "coding" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
6f2cf590cfcd313552a7612fae4c67a359ad2ce7
train_015046
2026-01-01T00:00:00
Agentic coding systems (plan→edit→test→reflect)
review
expert
Task: review Topic: Agentic coding systems (plan→edit→test→reflect) Difficulty: expert Language: C# Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "C#", "moe_experts": [ "security", "governance" ], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true }
05ac77711c79499ae41fb177f1d018a13563fa42
train_015047
2026-01-01T00:00:00
Code-specialized model families and scaling laws
explain
advanced
Task: explain Topic: Code-specialized model families and scaling laws Difficulty: advanced Language: Java Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Java", "moe_experts": [ "governance", "performance" ], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true }
67cfafe62fab2fa22675ffc2eb2dc3f0c5c1628c
train_015048
2026-01-01T00:00:00
Tool calling protocols and sandboxed execution
review
expert
Task: review Topic: Tool calling protocols and sandboxed execution Difficulty: expert Language: Java Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Java", "moe_experts": [ "governance", "agent" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
cd1db49fa51086f52e368e71278d7717d9eff71e
train_015049
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
data_pipeline
advanced
Task: data_pipeline Topic: SWE-bench real-repo evaluation and leakage control Difficulty: advanced Language: Python Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Python", "moe_experts": [ "research", "evaluation" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
be90dcc84e3f36cfc7704e1b3ef7644545831089
train_015050
2026-01-01T00:00:00
Dataset curation pipelines and quality attribution
explain
advanced
Task: explain Topic: Dataset curation pipelines and quality attribution Difficulty: advanced Language: Rust Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Rust", "moe_experts": [ "evaluation", "data" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
690989dfefb01ac012542e86bc3d5617a72b5480
train_015051
2026-01-01T00:00:00
Reasoning-first coding models and tunable deliberation
code
advanced
Task: code Topic: Reasoning-first coding models and tunable deliberation Difficulty: advanced Language: Bash Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Bash", "moe_experts": [ "research", "governance" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
6b118f34f3a85ecebfd0c434743054d00eb56a77
train_015052
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
review
advanced
Task: review Topic: SWE-bench real-repo evaluation and leakage control Difficulty: advanced Language: Go Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Go", "moe_experts": [ "data", "security" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
a8f167b2a66d9652cea1dc14230e289a4f157ee5
train_015053
2026-01-01T00:00:00
Agentic coding systems (plan→edit→test→reflect)
tool_trace
advanced
Task: tool_trace Topic: Agentic coding systems (plan→edit→test→reflect) Difficulty: advanced Language: Python Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "Python", "moe_experts": [ "agent", "governance" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
046519ef6cec61e47be541904a170e952ffca8bf
train_015054
2026-01-01T00:00:00
Reasoning-first coding models and tunable deliberation
data_pipeline
advanced
Task: data_pipeline Topic: Reasoning-first coding models and tunable deliberation Difficulty: advanced Language: Rust Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Rust", "moe_experts": [ "performance", "agent" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
45ada70b3538f0385104bcbda4b3f3ca8fefc352
train_015055
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
agent_loop
advanced
Task: agent_loop Topic: SWE-bench real-repo evaluation and leakage control Difficulty: advanced Language: SQL Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "SQL", "moe_experts": [ "research", "data" ], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true }
46bb19a840d62fbc35f3c39b692778b3d39aa73a
train_015056
2026-01-01T00:00:00
Failure engineering and resilience testing
compare
expert
Task: compare Topic: Failure engineering and resilience testing Difficulty: expert Language: Rust Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Rust", "moe_experts": [ "evaluation", "coding" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
bd2e8d705f09e9cd889039586ac6b4a005eba75b
train_015057
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
code
advanced
Task: code Topic: SWE-bench real-repo evaluation and leakage control Difficulty: advanced Language: Python Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Code: ```python def agent_loop(plan, edit, test, issue, budget=4): cost = 0 for _ in range(budget): p = plan(issue) patch = edi...
{ "language": "Python", "moe_experts": [ "coding", "performance" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
86edd581571f9c1cd4898db47fc1ea802d150a36
train_015058
2026-01-01T00:00:00
Latency, cost, throughput optimization
failure_analysis
advanced
Task: failure_analysis Topic: Latency, cost, throughput optimization Difficulty: advanced Language: Python Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Failure: regression detected Cause: missing edge case Fix: add guard + test
{ "language": "Python", "moe_experts": [ "security", "governance" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
ef73773ce8d321ef20d1f45370f456eef317240a
train_015059
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
code
expert
Task: code Topic: SWE-bench real-repo evaluation and leakage control Difficulty: expert Language: Rust Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Rust", "moe_experts": [ "coding", "governance" ], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true }
2a9568efa981f0ab927e0f2d766882e897f381fe
train_015060
2026-01-01T00:00:00
Self-grading and automated model critique
compare
expert
Task: compare Topic: Self-grading and automated model critique Difficulty: expert Language: Go Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Go", "moe_experts": [ "performance", "evaluation" ], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true }
b3c7564f745b9cd69cbb72c290aee29c8675f996
train_015061
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
explain
expert
Task: explain Topic: SWE-bench real-repo evaluation and leakage control Difficulty: expert Language: Bash Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Bash", "moe_experts": [ "data", "security" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
bd2fd5a21c03bf57a82461428def935bc9f8c8ba
train_015062
2026-01-01T00:00:00
Extended context, memory, and repo graphs
compare
advanced
Task: compare Topic: Extended context, memory, and repo graphs Difficulty: advanced Language: Go Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Go", "moe_experts": [ "performance", "data" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
2c844f109a72e96dd4d00725702893a7b6a1666f
train_015063
2026-01-01T00:00:00
Failure engineering and resilience testing
self_grade
advanced
Task: self_grade Topic: Failure engineering and resilience testing Difficulty: advanced Language: Go Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. { "score": 8, "confidence": 0.8, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "Go", "moe_experts": [ "coding", "agent" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
f611a7b0e277842ab270b0fde23da08e8d93f3c0
train_015064
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
agent_loop
advanced
Task: agent_loop Topic: SWE-bench real-repo evaluation and leakage control Difficulty: advanced Language: Python Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Python", "moe_experts": [ "coding", "evaluation" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
942453d70527173e2bd62dec865e6b58abb316b3
train_015065
2026-01-01T00:00:00
Self-grading and automated model critique
self_grade
advanced
Task: self_grade Topic: Self-grading and automated model critique Difficulty: advanced Language: Rust Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. { "score": 8, "confidence": 0.8, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "Rust", "moe_experts": [ "security", "agent" ], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true }
717a2602124e5d63ec1121b29ff1960d325047e4
train_015066
2026-01-01T00:00:00
Model merging, distillation, and continual learning
eval
expert
Task: eval Topic: Model merging, distillation, and continual learning Difficulty: expert Language: TypeScript Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Metrics: - pass@k - time-to-green - cost-per-fix - regression rate
{ "language": "TypeScript", "moe_experts": [ "agent", "security" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
445994dfb7a62391e6095cc0928b023e61f20804
train_015067
2026-01-01T00:00:00
Self-grading and automated model critique
explain
expert
Task: explain Topic: Self-grading and automated model critique Difficulty: expert Language: Rust Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Rust", "moe_experts": [ "coding", "agent" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
2c639cf4364db01cb1eacf5aa506a5ff04fceaaa
train_015068
2026-01-01T00:00:00
Model merging, distillation, and continual learning
failure_analysis
advanced
Task: failure_analysis Topic: Model merging, distillation, and continual learning Difficulty: advanced Language: Go Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Failure: regression detected Cause: missing edge case Fix: add guard + test
{ "language": "Go", "moe_experts": [ "agent", "coding" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
2b3ffb3cd7de0b1506e027b8268e1471a6576429
train_015069
2026-01-01T00:00:00
Latency, cost, throughput optimization
patch_diff
expert
Task: patch_diff Topic: Latency, cost, throughput optimization Difficulty: expert Language: SQL Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Patch: ```diff - result = a / b + if b == 0: + raise ValueError('division by zero') + result = a / b ```
{ "language": "SQL", "moe_experts": [ "security", "governance" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
fe41d604e3d5c2c4814d9b725852f831a26a2b43
train_015070
2026-01-01T00:00:00
Mixture-of-Experts (MoE) routing and sparsity
failure_analysis
expert
Task: failure_analysis Topic: Mixture-of-Experts (MoE) routing and sparsity Difficulty: expert Language: Bash Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Failure: regression detected Cause: missing edge case Fix: add guard + test
{ "language": "Bash", "moe_experts": [ "coding", "research" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
698804ee6110287cfdff4bee04787388b80ad193
train_015071
2026-01-01T00:00:00
Reasoning-first coding models and tunable deliberation
patch_diff
expert
Task: patch_diff Topic: Reasoning-first coding models and tunable deliberation Difficulty: expert Language: Python Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Patch: ```diff - result = a / b + if b == 0: + raise ValueError('division by zero') + result = a / b ```
{ "language": "Python", "moe_experts": [ "data", "governance" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
40df186b3b381791c14281616b7cd6d692217ca9
train_015072
2026-01-01T00:00:00
Tool calling protocols and sandboxed execution
compare
expert
Task: compare Topic: Tool calling protocols and sandboxed execution Difficulty: expert Language: SQL Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "SQL", "moe_experts": [ "coding", "research" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
7e25322ad52439477833d62f2e503dea9f2210af
train_015073
2026-01-01T00:00:00
Extended context, memory, and repo graphs
failure_analysis
advanced
Task: failure_analysis Topic: Extended context, memory, and repo graphs Difficulty: advanced Language: Java Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Failure: regression detected Cause: missing edge case Fix: add guard + test
{ "language": "Java", "moe_experts": [ "performance", "agent" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
dc34b2378f9b796e02cc1d6e1e780cb657b1224d
train_015074
2026-01-01T00:00:00
Mixture-of-Experts (MoE) routing and sparsity
eval
advanced
Task: eval Topic: Mixture-of-Experts (MoE) routing and sparsity Difficulty: advanced Language: SQL Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Metrics: - pass@k - time-to-green - cost-per-fix - regression rate
{ "language": "SQL", "moe_experts": [ "performance", "research" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
eca6d312da145f87f35062d2c14292b9fa1f967c
train_015075
2026-01-01T00:00:00
Governance, provenance, and audit logging
review
advanced
Task: review Topic: Governance, provenance, and audit logging Difficulty: advanced Language: SQL Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "SQL", "moe_experts": [ "research", "coding" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
bbbb0f6941681e246a104a25e5b118921bc211df
train_015076
2026-01-01T00:00:00
Code-specialized model families and scaling laws
agent_loop
expert
Task: agent_loop Topic: Code-specialized model families and scaling laws Difficulty: expert Language: Python Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Python", "moe_experts": [ "performance", "evaluation" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
cbff56cf82b7fde9c92ba331d0000fe8126f9e82
train_015077
2026-01-01T00:00:00
Model merging, distillation, and continual learning
agent_loop
advanced
Task: agent_loop Topic: Model merging, distillation, and continual learning Difficulty: advanced Language: Python Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Python", "moe_experts": [ "data", "performance" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
441286549af551f1a1ed1099af5178b059b2c047
train_015078
2026-01-01T00:00:00
Model merging, distillation, and continual learning
agent_loop
advanced
Task: agent_loop Topic: Model merging, distillation, and continual learning Difficulty: advanced Language: Python Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Python", "moe_experts": [ "security", "coding" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
95a4e85a405d61df531bbc34b9126059b33f8244
train_015079
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
self_grade
expert
Task: self_grade Topic: SWE-bench real-repo evaluation and leakage control Difficulty: expert Language: Bash Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. { "score": 6, "confidence": 0.6, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "Bash", "moe_experts": [ "data", "research" ], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true }
4cfc3a7b9ac2564315a784fd5f44d96215ecf5f9
train_015080
2026-01-01T00:00:00
Dataset curation pipelines and quality attribution
design
advanced
Task: design Topic: Dataset curation pipelines and quality attribution Difficulty: advanced Language: TypeScript Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "TypeScript", "moe_experts": [ "research", "data" ], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true }
459ae868a93d8126ff71e746923764048d82c38a
train_015081
2026-01-01T00:00:00
Multi-agent coordination and role separation
patch_diff
expert
Task: patch_diff Topic: Multi-agent coordination and role separation Difficulty: expert Language: Rust Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Patch: ```diff - result = a / b + if b == 0: + raise ValueError('division by zero') + result = a / b ```
{ "language": "Rust", "moe_experts": [ "coding", "security" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
83af418b8bed10fa5586feb69a2070dc1c04b3bf
train_015082
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
code
expert
Task: code Topic: SWE-bench real-repo evaluation and leakage control Difficulty: expert Language: Go Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Go", "moe_experts": [ "coding", "security" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
498a9240bba7b6ecb4aaf6352501003d1a4a330d
train_015083
2026-01-01T00:00:00
Tool calling protocols and sandboxed execution
tool_trace
advanced
Task: tool_trace Topic: Tool calling protocols and sandboxed execution Difficulty: advanced Language: Rust Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "Rust", "moe_experts": [ "research", "data" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
a5d72bc05e11895c82deefee8df4a9df2ec540c3
train_015084
2026-01-01T00:00:00
Dataset curation pipelines and quality attribution
review
advanced
Task: review Topic: Dataset curation pipelines and quality attribution Difficulty: advanced Language: JavaScript Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "JavaScript", "moe_experts": [ "security", "governance" ], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true }
827ac587ec6d47e1d4c56de18efe270c11b34290
train_015085
2026-01-01T00:00:00
Tool calling protocols and sandboxed execution
failure_analysis
expert
Task: failure_analysis Topic: Tool calling protocols and sandboxed execution Difficulty: expert Language: Go Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Failure: regression detected Cause: missing edge case Fix: add guard + test
{ "language": "Go", "moe_experts": [ "agent", "data" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
feefaae9b6c3d7885f36913f02ca8184a281db7b
train_015086
2026-01-01T00:00:00
Code-specialized model families and scaling laws
self_grade
advanced
Task: self_grade Topic: Code-specialized model families and scaling laws Difficulty: advanced Language: Java Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. { "score": 6, "confidence": 0.6, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "Java", "moe_experts": [ "data", "performance" ], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true }
4722d08f0c0c9f1c7963300174d937182df1a5f5
train_015087
2026-01-01T00:00:00
Secure code generation, red-teaming, and policy gates
self_grade
advanced
Task: self_grade Topic: Secure code generation, red-teaming, and policy gates Difficulty: advanced Language: Bash Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. { "score": 10, "confidence": 1.0, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "Bash", "moe_experts": [ "research", "performance" ], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true }
6a96855d55e01ce682bc38c38f67cdf79d11700c
train_015088
2026-01-01T00:00:00
Failure engineering and resilience testing
failure_analysis
expert
Task: failure_analysis Topic: Failure engineering and resilience testing Difficulty: expert Language: C# Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Failure: regression detected Cause: missing edge case Fix: add guard + test
{ "language": "C#", "moe_experts": [ "governance", "data" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
b9405b0805e54ed1c719387b1b6d3b0fbf097803
train_015089
2026-01-01T00:00:00
Mixture-of-Experts (MoE) routing and sparsity
explain
advanced
Task: explain Topic: Mixture-of-Experts (MoE) routing and sparsity Difficulty: advanced Language: C# Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "C#", "moe_experts": [ "governance", "data" ], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true }
3c87174a2e8ef2c212989d0e5dfb75e869e3fe97
train_015090
2026-01-01T00:00:00
Dataset curation pipelines and quality attribution
self_grade
expert
Task: self_grade Topic: Dataset curation pipelines and quality attribution Difficulty: expert Language: Rust Context: Research benchmark validation environment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. { "score": 9, "confidence": 0.9, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "Rust", "moe_experts": [ "security", "governance" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
b5f3e5cf2aa79035bec1907c6f96eb49a22c527c
train_015091
2026-01-01T00:00:00
Failure engineering and resilience testing
explain
expert
Task: explain Topic: Failure engineering and resilience testing Difficulty: expert Language: Go Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Go", "moe_experts": [ "evaluation", "coding" ], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true }
64c9fb7963a72a658e6cd3d3cbabe19cf0d52a1f
train_015092
2026-01-01T00:00:00
Mixture-of-Experts (MoE) routing and sparsity
design
expert
Task: design Topic: Mixture-of-Experts (MoE) routing and sparsity Difficulty: expert Language: Java Context: Offline air-gapped deployment Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Java", "moe_experts": [ "research", "security" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
84197167459f5ccc5e814579347770708b3a45c4
train_015093
2026-01-01T00:00:00
Failure engineering and resilience testing
compare
expert
Task: compare Topic: Failure engineering and resilience testing Difficulty: expert Language: Java Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Java", "moe_experts": [ "agent", "research" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
f8cc82ed8243e1ef31fcf0fd36c8c13ecb062c9f
train_015094
2026-01-01T00:00:00
Failure engineering and resilience testing
self_grade
expert
Task: self_grade Topic: Failure engineering and resilience testing Difficulty: expert Language: Bash Context: Regulated enterprise with audit requirements Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. { "score": 10, "confidence": 1.0, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "Bash", "moe_experts": [ "evaluation", "security" ], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true }
d3c5bb27b931df9b7bead7c8fd4e150c44909434
train_015095
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
failure_analysis
expert
Task: failure_analysis Topic: SWE-bench real-repo evaluation and leakage control Difficulty: expert Language: SQL Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Failure: regression detected Cause: missing edge case Fix: add guard + test
{ "language": "SQL", "moe_experts": [ "research", "governance" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
eda6f45d1b9b166c67442e8c55e2e389b445609d
train_015096
2026-01-01T00:00:00
Code-specialized model families and scaling laws
explain
expert
Task: explain Topic: Code-specialized model families and scaling laws Difficulty: expert Language: C# Context: Large monorepo with flaky tests and strict CI Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "C#", "moe_experts": [ "security", "evaluation" ], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true }
483aa43edad00b8df0814b6964a521184b87b695
train_015097
2026-01-01T00:00:00
Secure code generation, red-teaming, and policy gates
design
advanced
Task: design Topic: Secure code generation, red-teaming, and policy gates Difficulty: advanced Language: Python Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Expert guidance with risks, metrics, and governance controls.
{ "language": "Python", "moe_experts": [ "data", "agent" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
ce607899df184d88e278bec5b77ed00a3456ac24
train_015098
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
self_grade
expert
Task: self_grade Topic: SWE-bench real-repo evaluation and leakage control Difficulty: expert Language: JavaScript Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. { "score": 8, "confidence": 0.8, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "JavaScript", "moe_experts": [ "research", "agent" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
e301eab1ab8d6c171d69b6c0019c06d7ca4f2774
train_015099
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
eval
expert
Task: eval Topic: SWE-bench real-repo evaluation and leakage control Difficulty: expert Language: Bash Context: Latency-critical production API Produce frontier-grade, production-valid output.
Facts: - Modern AI coding success is measured by test pass rates, iteration cost, and reliability. - Agentic and multi-agent systems outperform single-pass generation. Metrics: - pass@k - time-to-green - cost-per-fix - regression rate
{ "language": "Bash", "moe_experts": [ "evaluation", "security" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
eff785af56a4385e33fb37529c1a861a17f846fa