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_038000
2026-01-01T00:00:00
Tool calling protocols and sandboxed execution
design
advanced
Task: design Topic: Tool calling protocols and sandboxed execution 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. Expert guidance with risks, metrics, and governance controls.
{ "language": "Bash", "moe_experts": [ "performance", "data" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
2313d27830bfcdecd3f855be90bff03fa62955d8
train_038001
2026-01-01T00:00:00
Reasoning-first coding models and tunable deliberation
data_pipeline
expert
Task: data_pipeline 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": [ "evaluation", "data" ], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true }
07a2af5e8b18a939d7ae2b54485017afbc65257d
train_038002
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: 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. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "TypeScript", "moe_experts": [ "governance", "evaluation" ], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true }
0e609322e07e53098effa126d38934300b0a06e8
train_038003
2026-01-01T00:00:00
Governance, provenance, and audit logging
design
expert
Task: design Topic: Governance, provenance, and audit logging 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", "agent" ], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true }
81ac9a92638657745501e6ed7a8c89fd017b528b
train_038004
2026-01-01T00:00:00
Latency, cost, throughput optimization
tool_trace
advanced
Task: tool_trace Topic: Latency, cost, throughput optimization Difficulty: advanced Language: Rust 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": "Rust", "moe_experts": [ "evaluation", "coding" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
8165940eebc3729b93f5affc2a88393e91641712
train_038005
2026-01-01T00:00:00
Agentic coding systems (plan→edit→test→reflect)
failure_analysis
expert
Task: failure_analysis 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. Failure: regression detected Cause: missing edge case Fix: add guard + test
{ "language": "Bash", "moe_experts": [ "coding", "research" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
456f7ec74579d7eba33b10548bd3cc358f249c85
train_038006
2026-01-01T00:00:00
Multi-agent coordination and role separation
review
advanced
Task: review Topic: Multi-agent coordination and role separation 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": [ "agent", "security" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
0ba07aa5c86f00bfeb9e2e74517fd07354a067e3
train_038007
2026-01-01T00:00:00
Multi-agent coordination and role separation
review
advanced
Task: review Topic: Multi-agent coordination and role separation 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": [ "evaluation", "data" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
5eb0b57e12c1c1e2ce2f5c3375f98ebd13d98abe
train_038008
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: 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": [ "coding", "research" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
44a8afe560771e710e3c5320f279bc1909af61a5
train_038009
2026-01-01T00:00:00
Code-specialized model families and scaling laws
code
expert
Task: code Topic: Code-specialized model families and scaling laws Difficulty: expert 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": [ "data", "security" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
65db213be565a4692ce940dc6db82dc431d1734c
train_038010
2026-01-01T00:00:00
Self-grading and automated model critique
tool_trace
advanced
Task: tool_trace Topic: Self-grading and automated model critique 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. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "C#", "moe_experts": [ "evaluation", "governance" ], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true }
e001a2f4c9446195168209942bb0a87e8a4fc39d
train_038011
2026-01-01T00:00:00
Mixture-of-Experts (MoE) routing and sparsity
compare
advanced
Task: compare Topic: Mixture-of-Experts (MoE) routing and sparsity 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. Expert guidance with risks, metrics, and governance controls.
{ "language": "JavaScript", "moe_experts": [ "agent", "research" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
8d20a7497aaa2edb4a4206ef083581f739f1d155
train_038012
2026-01-01T00:00:00
Mixture-of-Experts (MoE) routing and sparsity
self_grade
advanced
Task: self_grade Topic: Mixture-of-Experts (MoE) routing and sparsity 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. { "score": 8, "confidence": 0.8, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "Rust", "moe_experts": [ "security", "governance" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
67c4d9ef419aa976a6ef53fb2a786df6a960f2c8
train_038013
2026-01-01T00:00:00
Governance, provenance, and audit logging
tool_trace
expert
Task: tool_trace Topic: Governance, provenance, and audit logging 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. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "SQL", "moe_experts": [ "agent", "evaluation" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
195beffcd97e1b57aa813b2c5766ccec0abbf522
train_038014
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: 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. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "TypeScript", "moe_experts": [ "performance", "agent" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
cd0da55f720a19937ce71a79893878a6ba376b5f
train_038015
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: 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. { "score": 6, "confidence": 0.6, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "Python", "moe_experts": [ "governance", "security" ], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true }
2df2ea66ccd88419311464b45352a43d79323bdd
train_038016
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: 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": "Rust", "moe_experts": [ "research", "coding" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
377d8df923cec92532b8ff8a0335768ab4c135e5
train_038017
2026-01-01T00:00:00
Extended context, memory, and repo graphs
review
expert
Task: review Topic: Extended context, memory, and repo graphs Difficulty: expert 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. Expert guidance with risks, metrics, and governance controls.
{ "language": "C#", "moe_experts": [ "data", "performance" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
48da2302a2b24d2e8f827af8612456dbf9ba5564
train_038018
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: 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": [ "evaluation", "data" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
852c0c1adc0bda9811a56f045ed7d40dcc0171be
train_038019
2026-01-01T00:00:00
Multi-agent coordination and role separation
tool_trace
advanced
Task: tool_trace Topic: Multi-agent coordination and role separation 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. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "Go", "moe_experts": [ "coding", "governance" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
b88ad6ae525d1721e46b76230c1c087675af08e5
train_038020
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: 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": [ "security", "evaluation" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
44005b060bb8add469e47563fd57f9a315656a65
train_038021
2026-01-01T00:00:00
Secure code generation, red-teaming, and policy gates
compare
advanced
Task: compare Topic: Secure code generation, red-teaming, and policy gates 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": [ "coding", "governance" ], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true }
30c63915e2a431a60049b78cee673e5603b1c33b
train_038022
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: 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. { "score": 7, "confidence": 0.7, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "TypeScript", "moe_experts": [ "data", "performance" ], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true }
300f9d1bc5865fd3561d0f513a306c925bdd27e9
train_038023
2026-01-01T00:00:00
Tool calling protocols and sandboxed execution
data_pipeline
advanced
Task: data_pipeline Topic: Tool calling protocols and sandboxed execution 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": [ "governance", "performance" ], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true }
12d72501c3b2948ac2d225a2f704fcd639186f7d
train_038024
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": [ "research", "security" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
0073d9f517528c565ec725ac49f21a6e11dd687d
train_038025
2026-01-01T00:00:00
Multi-agent coordination and role separation
compare
expert
Task: compare Topic: Multi-agent coordination and role separation 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. Expert guidance with risks, metrics, and governance controls.
{ "language": "SQL", "moe_experts": [ "data", "performance" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
eaf09b41539299b2db2d8353f22670b6fd5af487
train_038026
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: 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. Patch: ```diff - result = a / b + if b == 0: + raise ValueError('division by zero') + result = a / b ```
{ "language": "JavaScript", "moe_experts": [ "agent", "data" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
8dd6650f3003bfc36d31e4081042889483f52468
train_038027
2026-01-01T00:00:00
Code-specialized model families and scaling laws
design
expert
Task: design Topic: Code-specialized model families and scaling laws Difficulty: expert Language: Java 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": "Java", "moe_experts": [ "coding", "agent" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
a4904a44b1a1cc1f1c4c3f2c7cd3ed6c6bee9bc8
train_038028
2026-01-01T00:00:00
Self-grading and automated model critique
patch_diff
advanced
Task: patch_diff Topic: Self-grading and automated model critique 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. Patch: ```diff - result = a / b + if b == 0: + raise ValueError('division by zero') + result = a / b ```
{ "language": "SQL", "moe_experts": [ "coding", "agent" ], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true }
02561af755791af61118326abc59ef9c13e56643
train_038029
2026-01-01T00:00:00
Agentic coding systems (plan→edit→test→reflect)
eval
expert
Task: eval Topic: Agentic coding systems (plan→edit→test→reflect) Difficulty: expert 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. Metrics: - pass@k - time-to-green - cost-per-fix - regression rate
{ "language": "Python", "moe_experts": [ "security", "performance" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
fb0db58d102eaae5c37069af86f67dfe498e2e71
train_038030
2026-01-01T00:00:00
Multi-agent coordination and role separation
design
advanced
Task: design Topic: Multi-agent coordination and role separation 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": [ "performance", "evaluation" ], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true }
e36512f17f8586d789576dfe1d86eccc38e59da4
train_038031
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
data_pipeline
expert
Task: data_pipeline Topic: SWE-bench real-repo evaluation and leakage control Difficulty: expert 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": [ "coding", "performance" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
dc68260f97746fc7b4697be1b5442ec10be85176
train_038032
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: 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": [ "data", "research" ], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true }
13b580eb16f6234c05cb186d2aa2f51ea07e5d6f
train_038033
2026-01-01T00:00:00
Governance, provenance, and audit logging
review
expert
Task: review Topic: Governance, provenance, and audit logging 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": [ "security", "research" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
0bd479ec7248877fe5cfed84f2a4a43387b24d96
train_038034
2026-01-01T00:00:00
Secure code generation, red-teaming, and policy gates
explain
expert
Task: explain Topic: Secure code generation, red-teaming, and policy gates Difficulty: expert 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": [ "evaluation", "governance" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
160463cee8d58546990e1076072b806b3d14e76b
train_038035
2026-01-01T00:00:00
Multi-agent coordination and role separation
tool_trace
advanced
Task: tool_trace Topic: Multi-agent coordination and role separation 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. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "TypeScript", "moe_experts": [ "security", "research" ], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true }
8fb3aae6fa2ea3cb52d3d8825a4584dbc3f2bcea
train_038036
2026-01-01T00:00:00
Agentic coding systems (plan→edit→test→reflect)
code
advanced
Task: code Topic: Agentic coding systems (plan→edit→test→reflect) 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": 4, "latency_ms_target": 50, "audit_required": true }
80cb6d32a77f68d7ad039a2cd4b2c0da5a44e566
train_038037
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: 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": [ "agent", "data" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
89709c1afb1d85db53f088dd5361634adf11f42f
train_038038
2026-01-01T00:00:00
Latency, cost, throughput optimization
tool_trace
expert
Task: tool_trace 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. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "SQL", "moe_experts": [ "research", "data" ], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true }
8fb0a33b3879fcc4d65c562a5f7a8890e85e7092
train_038039
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: 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. Expert guidance with risks, metrics, and governance controls.
{ "language": "Bash", "moe_experts": [ "agent", "research" ], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true }
61faf614c8564411fa0d117428efbfcce2f37b1f
train_038040
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
tool_trace
advanced
Task: tool_trace Topic: SWE-bench real-repo evaluation and leakage control 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. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "Rust", "moe_experts": [ "coding", "security" ], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true }
d42e103fd606a8408d4730580023dbc18080cdbe
train_038041
2026-01-01T00:00:00
Reasoning-first coding models and tunable deliberation
review
expert
Task: review Topic: Reasoning-first coding models and tunable deliberation 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": [ "agent", "data" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
2c40bf61ee35834cc8c94a399da7390d2b8cc904
train_038042
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: 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": [ "evaluation", "data" ], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true }
2ad406382a090ae31a6591f49240b0ea2616f604
train_038043
2026-01-01T00:00:00
Dataset curation pipelines and quality attribution
patch_diff
advanced
Task: patch_diff Topic: Dataset curation pipelines and quality attribution 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. Patch: ```diff - result = a / b + if b == 0: + raise ValueError('division by zero') + result = a / b ```
{ "language": "Bash", "moe_experts": [ "governance", "research" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
aa709c61609590a9cc349cd02bbd72f2908cfbfd
train_038044
2026-01-01T00:00:00
Code-specialized model families and scaling laws
patch_diff
advanced
Task: patch_diff Topic: Code-specialized model families and scaling laws 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. Patch: ```diff - result = a / b + if b == 0: + raise ValueError('division by zero') + result = a / b ```
{ "language": "TypeScript", "moe_experts": [ "research", "performance" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
1418c71953c6dd0cb8550c4c86fc4ac99835f0e1
train_038045
2026-01-01T00:00:00
Reasoning-first coding models and tunable deliberation
tool_trace
expert
Task: tool_trace Topic: Reasoning-first coding models and tunable deliberation 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. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "Python", "moe_experts": [ "performance", "agent" ], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true }
b8d4f02bc144cce05078098aaaf4d456f9eb0784
train_038046
2026-01-01T00:00:00
Failure engineering and resilience testing
agent_loop
expert
Task: agent_loop Topic: Failure engineering and resilience testing Difficulty: expert 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": [ "performance", "data" ], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true }
b914a6fc0b8bc589011e30bdca7749da727182c3
train_038047
2026-01-01T00:00:00
Multi-agent coordination and role separation
self_grade
expert
Task: self_grade Topic: Multi-agent coordination and role separation 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. { "score": 7, "confidence": 0.7, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "SQL", "moe_experts": [ "performance", "evaluation" ], "cost_budget": 4, "latency_ms_target": 200, "audit_required": true }
27ed7080ae70aabe878a08d6f960df7fd21d77fb
train_038048
2026-01-01T00:00:00
Failure engineering and resilience testing
code
advanced
Task: code Topic: Failure engineering and resilience testing 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", "security" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
1b66c4e0fff2f015d4e0158659712be4b7152acd
train_038049
2026-01-01T00:00:00
Mixture-of-Experts (MoE) routing and sparsity
code
advanced
Task: code Topic: Mixture-of-Experts (MoE) routing and sparsity Difficulty: advanced 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": [ "security", "evaluation" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
1b1390ac273f03a252e8d2700a0cf2dab7c6671c
train_038050
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
patch_diff
expert
Task: patch_diff Topic: SWE-bench real-repo evaluation and leakage control Difficulty: expert 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. Patch: ```diff - result = a / b + if b == 0: + raise ValueError('division by zero') + result = a / b ```
{ "language": "Rust", "moe_experts": [ "agent", "coding" ], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true }
e8dfd42424d11d9e324683a2e04efc6384811bcb
train_038051
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
review
expert
Task: review Topic: SWE-bench real-repo evaluation and leakage control Difficulty: expert 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": [ "coding", "agent" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
45bb948b381a655dabef03521f4e689d1f04ab12
train_038052
2026-01-01T00:00:00
Dataset curation pipelines and quality attribution
review
expert
Task: review Topic: Dataset curation pipelines and quality attribution Difficulty: expert 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": [ "coding", "data" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
65b4ad841c594f6adb032d8ea5bcf079ec325c6c
train_038053
2026-01-01T00:00:00
Multi-agent coordination and role separation
code
advanced
Task: code Topic: Multi-agent coordination and role separation Difficulty: advanced 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": [ "governance", "data" ], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true }
41da8c97baffc49d4bc7032e1a8178110a586475
train_038054
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: 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": [ "agent", "governance" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
ce6caf0ad86798d0e58db653d1f4e21d96801f90
train_038055
2026-01-01T00:00:00
Tool calling protocols and sandboxed execution
data_pipeline
advanced
Task: data_pipeline Topic: Tool calling protocols and sandboxed execution 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": [ "performance", "coding" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
fc60ed67bc183c310c64d794e4d2379fec69cdb4
train_038056
2026-01-01T00:00:00
Tool calling protocols and sandboxed execution
self_grade
expert
Task: self_grade Topic: Tool calling protocols and sandboxed execution 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": 8, "confidence": 0.8, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "Bash", "moe_experts": [ "coding", "governance" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
54529c2f6ba2055c3020e03b62f2ba9b17c46b61
train_038057
2026-01-01T00:00:00
Code-specialized model families and scaling laws
data_pipeline
advanced
Task: data_pipeline Topic: Code-specialized model families and scaling laws 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", "governance" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
14003e8d5e9eb98ebe68885a4ffc1b299df9d0fa
train_038058
2026-01-01T00:00:00
Governance, provenance, and audit logging
eval
expert
Task: eval Topic: Governance, provenance, and audit logging 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. Metrics: - pass@k - time-to-green - cost-per-fix - regression rate
{ "language": "Java", "moe_experts": [ "data", "performance" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
8dea89501e07ee3a81fcf324dd03ceb70b526dcd
train_038059
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: 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": [ "evaluation", "coding" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
5fbcfd5d4ef79cfbbd52c05e9fe716920acf2392
train_038060
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: 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": [ "security", "governance" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
539337e37b40c567588b5fbee2537481d57093e2
train_038061
2026-01-01T00:00:00
Failure engineering and resilience testing
explain
expert
Task: explain Topic: Failure engineering and resilience testing Difficulty: expert 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": [ "governance", "security" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
448b29b035826ea187eabcb57b16751ff11f94f5
train_038062
2026-01-01T00:00:00
Agentic coding systems (plan→edit→test→reflect)
failure_analysis
advanced
Task: failure_analysis Topic: Agentic coding systems (plan→edit→test→reflect) 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. Failure: regression detected Cause: missing edge case Fix: add guard + test
{ "language": "TypeScript", "moe_experts": [ "data", "security" ], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true }
171579cfce59abe5786fbd7e983226ddf647a9c6
train_038063
2026-01-01T00:00:00
Governance, provenance, and audit logging
explain
advanced
Task: explain Topic: Governance, provenance, and audit logging Difficulty: advanced Language: Rust 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": "Rust", "moe_experts": [ "evaluation", "research" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
23e91dd97ca412201230ece69adf651c83c29682
train_038064
2026-01-01T00:00:00
Secure code generation, red-teaming, and policy gates
design
expert
Task: design Topic: Secure code generation, red-teaming, and policy gates 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": [ "data", "security" ], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true }
20bae42e9ff16d92cb1c1c5baeafb1a9c866f309
train_038065
2026-01-01T00:00:00
Reasoning-first coding models and tunable deliberation
failure_analysis
expert
Task: failure_analysis Topic: Reasoning-first coding models and tunable deliberation 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. Failure: regression detected Cause: missing edge case Fix: add guard + test
{ "language": "Java", "moe_experts": [ "data", "security" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
c8123279a6f5ce0c99d582409d6286995b34762c
train_038066
2026-01-01T00:00:00
Latency, cost, throughput optimization
compare
advanced
Task: compare Topic: Latency, cost, throughput optimization 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": [ "performance", "agent" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
4178464ee52a77f5d904e7bd478bde6f20f5f649
train_038067
2026-01-01T00:00:00
Multi-agent coordination and role separation
self_grade
advanced
Task: self_grade Topic: Multi-agent coordination and role separation 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. { "score": 10, "confidence": 1.0, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "SQL", "moe_experts": [ "coding", "performance" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
bc123bc4d9b22d24065812953ae25b8053a88229
train_038068
2026-01-01T00:00:00
Secure code generation, red-teaming, and policy gates
compare
advanced
Task: compare Topic: Secure code generation, red-teaming, and policy gates Difficulty: advanced Language: Rust 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": "Rust", "moe_experts": [ "performance", "security" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
569a48afbf4fe97430035faf4959b36b40ba6dfd
train_038069
2026-01-01T00:00:00
Extended context, memory, and repo graphs
explain
expert
Task: explain Topic: Extended context, memory, and repo graphs 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": [ "coding", "performance" ], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true }
631d5965abb8d840258fc5b03b07bd3558eb2f01
train_038070
2026-01-01T00:00:00
Dataset curation pipelines and quality attribution
tool_trace
expert
Task: tool_trace Topic: Dataset curation pipelines and quality attribution Difficulty: expert 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. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "Bash", "moe_experts": [ "security", "data" ], "cost_budget": 2, "latency_ms_target": 50, "audit_required": true }
cc6e3f41dd8b5440e60d965a7111de362666beb4
train_038071
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: 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": [ "security", "coding" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
501fdb9574ced461ae8afddc7df576a1abaf981c
train_038072
2026-01-01T00:00:00
Self-grading and automated model critique
failure_analysis
expert
Task: failure_analysis Topic: Self-grading and automated model critique 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. Failure: regression detected Cause: missing edge case Fix: add guard + test
{ "language": "Java", "moe_experts": [ "coding", "security" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
6d48f9391d5823d521b5d254cb72b0f2e55b884d
train_038073
2026-01-01T00:00:00
Tool calling protocols and sandboxed execution
code
expert
Task: code Topic: Tool calling protocols and sandboxed execution Difficulty: expert 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", "security" ], "cost_budget": 3, "latency_ms_target": 100, "audit_required": true }
253b5856af155301401f711904bdca6f39b8afe7
train_038074
2026-01-01T00:00:00
Dataset curation pipelines and quality attribution
review
expert
Task: review Topic: Dataset curation pipelines and quality attribution 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": [ "performance", "coding" ], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true }
d8b858fcade573d9bbb63a963dbfc78b86427b52
train_038075
2026-01-01T00:00:00
Secure code generation, red-teaming, and policy gates
code
advanced
Task: code Topic: Secure code generation, red-teaming, and policy gates 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": [ "governance", "performance" ], "cost_budget": 3, "latency_ms_target": 50, "audit_required": true }
05989c31f40a7601c220fc55940ba073b684ec9c
train_038076
2026-01-01T00:00:00
Failure engineering and resilience testing
design
expert
Task: design Topic: Failure engineering and resilience testing Difficulty: expert 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": [ "coding", "security" ], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true }
ec4a3320cc836c3705f41075d4f60b0b8a15ce3e
train_038077
2026-01-01T00:00:00
Model merging, distillation, and continual learning
code
expert
Task: code Topic: Model merging, distillation, and continual learning Difficulty: expert 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": [ "evaluation", "coding" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
c9979df3daf8d79efa475341e79401469fb7b699
train_038078
2026-01-01T00:00:00
Dataset curation pipelines and quality attribution
eval
expert
Task: eval Topic: Dataset curation pipelines and quality attribution Difficulty: expert 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. Metrics: - pass@k - time-to-green - cost-per-fix - regression rate
{ "language": "TypeScript", "moe_experts": [ "security", "governance" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
4b75d31c3d22775b6e33d685e953ebc63bc1fd9b
train_038079
2026-01-01T00:00:00
SWE-bench real-repo evaluation and leakage control
tool_trace
expert
Task: tool_trace 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. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "Bash", "moe_experts": [ "security", "governance" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
c4023bda6e0b9090701791164e600733e8f434be
train_038080
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: 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": [ "evaluation", "data" ], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true }
5a933a8831506f9f7e8aa756b369cde57e368915
train_038081
2026-01-01T00:00:00
Failure engineering and resilience testing
explain
advanced
Task: explain Topic: Failure engineering and resilience testing 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": [ "security", "evaluation" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
277189aeea628e93abb491910e2020c1812f4909
train_038082
2026-01-01T00:00:00
Extended context, memory, and repo graphs
review
expert
Task: review Topic: Extended context, memory, and repo graphs Difficulty: expert Language: Rust 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": "Rust", "moe_experts": [ "coding", "evaluation" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
3ded3d6757f871d8be45a900bea9d9f63658fac4
train_038083
2026-01-01T00:00:00
Self-grading and automated model critique
data_pipeline
expert
Task: data_pipeline Topic: Self-grading and automated model critique 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. Expert guidance with risks, metrics, and governance controls.
{ "language": "SQL", "moe_experts": [ "data", "agent" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
4702d42ad55190a02a872f2c36571d009af5a4bb
train_038084
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: 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": [ "governance", "security" ], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true }
3fbbdac39644e5186ef1131ef21cad08332d196e
train_038085
2026-01-01T00:00:00
Reasoning-first coding models and tunable deliberation
tool_trace
expert
Task: tool_trace Topic: Reasoning-first coding models and tunable deliberation 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. [ { "tool": "repo_checkout", "args": { "commit": "abc123" } }, { "tool": "run_tests", "args": { "suite": "unit" ...
{ "language": "C#", "moe_experts": [ "coding", "governance" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
780227552b2d2f617100f1c5f7ec2115a80328b0
train_038086
2026-01-01T00:00:00
Mixture-of-Experts (MoE) routing and sparsity
code
expert
Task: code Topic: Mixture-of-Experts (MoE) routing and sparsity Difficulty: expert 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. Expert guidance with risks, metrics, and governance controls.
{ "language": "C#", "moe_experts": [ "coding", "governance" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
38c547184f396e1219360e5e04e16122a1578f54
train_038087
2026-01-01T00:00:00
Mixture-of-Experts (MoE) routing and sparsity
data_pipeline
advanced
Task: data_pipeline Topic: Mixture-of-Experts (MoE) routing and sparsity 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": [ "evaluation", "coding" ], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true }
b35046027cb2275c4d681641020b85b1c7216140
train_038088
2026-01-01T00:00:00
Mixture-of-Experts (MoE) routing and sparsity
patch_diff
advanced
Task: patch_diff Topic: Mixture-of-Experts (MoE) routing and sparsity 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. Patch: ```diff - result = a / b + if b == 0: + raise ValueError('division by zero') + result = a / b ```
{ "language": "Rust", "moe_experts": [ "coding", "data" ], "cost_budget": 5, "latency_ms_target": 50, "audit_required": true }
8be4de57c505b41e69e1e7e06bf7c51725d1668d
train_038089
2026-01-01T00:00:00
Dataset curation pipelines and quality attribution
patch_diff
expert
Task: patch_diff Topic: Dataset curation pipelines and quality attribution 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. Patch: ```diff - result = a / b + if b == 0: + raise ValueError('division by zero') + result = a / b ```
{ "language": "Bash", "moe_experts": [ "governance", "performance" ], "cost_budget": 5, "latency_ms_target": 200, "audit_required": true }
4476809e6bbf860c673e245366d4f455967a9875
train_038090
2026-01-01T00:00:00
Model merging, distillation, and continual learning
explain
expert
Task: explain Topic: Model merging, distillation, and continual learning Difficulty: expert 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": [ "security", "research" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
87785d212ff2d46fee7308e60f09ff2e2802a25f
train_038091
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: 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", "research" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
f269ad566856cf4447a2ffadfdc258677d70170d
train_038092
2026-01-01T00:00:00
Tool calling protocols and sandboxed execution
eval
expert
Task: eval 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. Metrics: - pass@k - time-to-green - cost-per-fix - regression rate
{ "language": "Java", "moe_experts": [ "evaluation", "performance" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
a833ef6c370ba2868014b667f04d7bda55acb56c
train_038093
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: Java 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": "Java", "moe_experts": [ "performance", "agent" ], "cost_budget": 4, "latency_ms_target": 100, "audit_required": true }
e3798ffff1addacbf4d969261f8d1393b3a73cb0
train_038094
2026-01-01T00:00:00
Extended context, memory, and repo graphs
self_grade
advanced
Task: self_grade Topic: Extended context, memory, and repo graphs 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. { "score": 8, "confidence": 0.8, "notes": "Passed tests; minimal diff; no security flags." }
{ "language": "TypeScript", "moe_experts": [ "security", "coding" ], "cost_budget": 2, "latency_ms_target": 200, "audit_required": true }
b1dff6f8c3e35b0f1b56c547bfc643e92e2aec7c
train_038095
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: 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. Patch: ```diff - result = a / b + if b == 0: + raise ValueError('division by zero') + result = a / b ```
{ "language": "Bash", "moe_experts": [ "research", "evaluation" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
6da902051c46228746d1a42abe973e19c05a8e25
train_038096
2026-01-01T00:00:00
Code-specialized model families and scaling laws
code
expert
Task: code Topic: Code-specialized model families and scaling laws 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": [ "agent", "data" ], "cost_budget": 4, "latency_ms_target": 50, "audit_required": true }
ac59ff758f563480600e2d00879c89367ed5e5d6
train_038097
2026-01-01T00:00:00
Dataset curation pipelines and quality attribution
patch_diff
expert
Task: patch_diff Topic: Dataset curation pipelines and quality attribution Difficulty: expert 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. Patch: ```diff - result = a / b + if b == 0: + raise ValueError('division by zero') + result = a / b ```
{ "language": "C#", "moe_experts": [ "research", "evaluation" ], "cost_budget": 2, "latency_ms_target": 100, "audit_required": true }
af9bbb0f5db31671a7b52f4f304822f528050826
train_038098
2026-01-01T00:00:00
Tool calling protocols and sandboxed execution
explain
advanced
Task: explain Topic: Tool calling protocols and sandboxed execution 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": [ "performance", "agent" ], "cost_budget": 3, "latency_ms_target": 200, "audit_required": true }
2bf84414f60b30dca744f770bfb5d92794761dd2
train_038099
2026-01-01T00:00:00
Model merging, distillation, and continual learning
patch_diff
advanced
Task: patch_diff Topic: Model merging, distillation, and continual learning 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. Patch: ```diff - result = a / b + if b == 0: + raise ValueError('division by zero') + result = a / b ```
{ "language": "SQL", "moe_experts": [ "security", "agent" ], "cost_budget": 5, "latency_ms_target": 100, "audit_required": true }
ad808461a87edd0a486e915749a623b67cc149e8