id
stringlengths
11
11
created
timestamp[s]date
2026-01-01 00:00:00
2026-01-01 00:00:00
topic
stringclasses
14 values
task_type
stringclasses
10 values
difficulty
stringclasses
3 values
instruction
stringlengths
189
248
input
stringclasses
1 value
output
stringclasses
9 values
reasoning_steps
listlengths
0
5
metadata
dict
hash
stringlengths
40
40
train_21000
2026-01-01T00:00:00
Latency, cost, and reliability optimization
eval
advanced
Task: eval Topic: Latency, cost, and reliability optimization Difficulty: advanced Target language: Python Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Eval: pass@k, time-to-green, regressions, diff size
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Python", "developer_needs": [ "documentation", "governance", "repo_scale_reasoning", "tests_are_truth" ], "moe_experts": [ "evaluation_expert", "coding_expert" ], "governance": { "audit_required": true, "tests_required": true } }
5ba9a1d300bed3758be39b89b9b2312c343b516e
train_21001
2026-01-01T00:00:00
Model merging, distillation, and continued pretraining
code
intermediate
Task: code Topic: Model merging, distillation, and continued pretraining Difficulty: intermediate Target language: JavaScript Context: Regulated environment requiring audit trails. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "JavaScript", "developer_needs": [ "governance", "tooling", "evaluation_metrics", "security_gates" ], "moe_experts": [ "agentic_systems_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
2c99db5669b1592b572f35ee83591b1d31e1313e
train_21002
2026-01-01T00:00:00
Code-specialized model families and sizing tradeoffs
patch_diff
advanced
Task: patch_diff Topic: Code-specialized model families and sizing tradeoffs Difficulty: advanced Target language: SQL Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Patch (diff-style): ```diff - if x == 0: - return 1/x + if x == 0: + raise ValueError('division by zero') ``` Acceptance: - Tests pass - No new regressions
[]
{ "target_language": "SQL", "developer_needs": [ "ci_integration", "evaluation_metrics", "cost_latency_tradeoffs", "tests_are_truth" ], "moe_experts": [ "evaluation_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
bb3006328eae7b2f78bae436f47253e261decaa0
train_21003
2026-01-01T00:00:00
Secure code generation and policy gates
eval
expert
Task: eval Topic: Secure code generation and policy gates Difficulty: expert Target language: Python Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Eval: pass@k, time-to-green, regressions, diff size
[]
{ "target_language": "Python", "developer_needs": [ "auditability", "documentation", "cost_latency_tradeoffs", "governance" ], "moe_experts": [ "data_curation_expert", "performance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
b29b7103fe131a8b15e4f54f12c8e7f1f22c5a35
train_21004
2026-01-01T00:00:00
Dataset curation pipelines (filter, dedupe, quality)
eval
advanced
Task: eval Topic: Dataset curation pipelines (filter, dedupe, quality) Difficulty: advanced Target language: Java Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Eval: pass@k, time-to-green, regressions, diff size
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Java", "developer_needs": [ "tooling", "ci_integration", "reproducibility", "evaluation_metrics" ], "moe_experts": [ "data_curation_expert", "coding_expert" ], "governance": { "audit_required": true, "tests_required": true } }
e9827c06b35bd4161421a911040357fac51ca557
train_21005
2026-01-01T00:00:00
Dataset curation pipelines (filter, dedupe, quality)
data_pipeline
expert
Task: data_pipeline Topic: Dataset curation pipelines (filter, dedupe, quality) Difficulty: expert Target language: Python Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Pipeline: Ingest → Normalize → Filter → Dedupe → Quality → Mix → Audit
[]
{ "target_language": "Python", "developer_needs": [ "evaluation_metrics", "repo_scale_reasoning", "auditability", "tests_are_truth" ], "moe_experts": [ "performance_expert", "evaluation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
895992cd899766db8973f3889171cc8dd5dd9772
train_21006
2026-01-01T00:00:00
Governance, provenance, and licensing for code data
explain
expert
Task: explain Topic: Governance, provenance, and licensing for code data Difficulty: expert Target language: Java Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Java", "developer_needs": [ "tooling", "cost_latency_tradeoffs", "ci_integration", "security_gates" ], "moe_experts": [ "coding_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
b36c4c254a7041db5d9943567a21f6f3608fa95e
train_21007
2026-01-01T00:00:00
Latency, cost, and reliability optimization
design
advanced
Task: design Topic: Latency, cost, and reliability optimization Difficulty: advanced Target language: TypeScript Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "TypeScript", "developer_needs": [ "reproducibility", "evaluation_metrics", "ci_integration", "repo_scale_reasoning" ], "moe_experts": [ "performance_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
e457350b6cee52c900f188376580b4a012add9fd
train_21008
2026-01-01T00:00:00
Model merging, distillation, and continued pretraining
design
expert
Task: design Topic: Model merging, distillation, and continued pretraining Difficulty: expert Target language: Bash Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "Bash", "developer_needs": [ "tooling", "cost_latency_tradeoffs", "reproducibility", "evaluation_metrics" ], "moe_experts": [ "agentic_systems_expert", "coding_expert" ], "governance": { "audit_required": true, "tests_required": true } }
9399c5c78c849c1944c4c64faa64d158d6802d66
train_21009
2026-01-01T00:00:00
Latency, cost, and reliability optimization
agent_loop
advanced
Task: agent_loop Topic: Latency, cost, and reliability optimization Difficulty: advanced Target language: Bash Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Loop: Plan → Edit → Test → Reflect → Human gate
[]
{ "target_language": "Bash", "developer_needs": [ "governance", "cost_latency_tradeoffs", "documentation", "tooling" ], "moe_experts": [ "governance_expert", "coding_expert" ], "governance": { "audit_required": true, "tests_required": true } }
f711b09e61504412f9c32ce18950264ab67b8cd3
train_21010
2026-01-01T00:00:00
Governance, provenance, and licensing for code data
code
intermediate
Task: code Topic: Governance, provenance, and licensing for code data Difficulty: intermediate Target language: JavaScript Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "JavaScript", "developer_needs": [ "documentation", "repo_scale_reasoning", "reproducibility", "cost_latency_tradeoffs" ], "moe_experts": [ "security_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
d9768452fbb74886295779a4a8f0ba320fd2aeea
train_21011
2026-01-01T00:00:00
Reasoning-first coding models and tunable deliberation
review
intermediate
Task: review Topic: Reasoning-first coding models and tunable deliberation Difficulty: intermediate Target language: Python Context: Regulated environment requiring audit trails. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[]
{ "target_language": "Python", "developer_needs": [ "reproducibility", "evaluation_metrics", "auditability", "tooling" ], "moe_experts": [ "agentic_systems_expert", "governance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
8bf5e786a5c18d4b47b282b8960a4da1312c9929
train_21012
2026-01-01T00:00:00
Model merging, distillation, and continued pretraining
code
advanced
Task: code Topic: Model merging, distillation, and continued pretraining Difficulty: advanced Target language: SQL Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "SQL", "developer_needs": [ "auditability", "ci_integration", "cost_latency_tradeoffs", "documentation" ], "moe_experts": [ "performance_expert", "coding_expert" ], "governance": { "audit_required": true, "tests_required": true } }
20e7f57364324a0cf3e50c7b55b4a4881271d050
train_21013
2026-01-01T00:00:00
Governance, provenance, and licensing for code data
design
intermediate
Task: design Topic: Governance, provenance, and licensing for code data Difficulty: intermediate Target language: JavaScript Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "JavaScript", "developer_needs": [ "tests_are_truth", "security_gates", "tooling", "documentation" ], "moe_experts": [ "governance_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
fd31edc6ef37f24809912ae4b93637d6a6c14eba
train_21014
2026-01-01T00:00:00
Dataset curation pipelines (filter, dedupe, quality)
explain
expert
Task: explain Topic: Dataset curation pipelines (filter, dedupe, quality) Difficulty: expert Target language: SQL Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "SQL", "developer_needs": [ "cost_latency_tradeoffs", "governance", "tooling", "tests_are_truth" ], "moe_experts": [ "coding_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
a6a7e65500adef77f9c285853e553a63fab5d8a9
train_21015
2026-01-01T00:00:00
Mixture-of-Experts (MoE) for code
patch_diff
intermediate
Task: patch_diff Topic: Mixture-of-Experts (MoE) for code Difficulty: intermediate Target language: Python Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Patch (diff-style): ```diff - if x == 0: - return 1/x + if x == 0: + raise ValueError('division by zero') ``` Acceptance: - Tests pass - No new regressions
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Python", "developer_needs": [ "auditability", "documentation", "reproducibility", "tests_are_truth" ], "moe_experts": [ "performance_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
af1e5559c490a3b47c721305cbf71a3a066709ee
train_21016
2026-01-01T00:00:00
Latency, cost, and reliability optimization
design
intermediate
Task: design Topic: Latency, cost, and reliability optimization Difficulty: intermediate Target language: SQL Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "SQL", "developer_needs": [ "governance", "tests_are_truth", "security_gates", "ci_integration" ], "moe_experts": [ "data_curation_expert", "evaluation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
60ca990b73055c4879b2ccbd11bebe2b5de09844
train_21017
2026-01-01T00:00:00
Secure code generation and policy gates
code
intermediate
Task: code Topic: Secure code generation and policy gates Difficulty: intermediate Target language: Bash Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "Bash", "developer_needs": [ "auditability", "documentation", "governance", "evaluation_metrics" ], "moe_experts": [ "evaluation_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
01118ef55812ff1bc9065e055d94af05fdfe6546
train_21018
2026-01-01T00:00:00
Secure code generation and policy gates
failure_analysis
advanced
Task: failure_analysis Topic: Secure code generation and policy gates Difficulty: advanced Target language: Go Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Failure: - Initial patch broke edge case Reflection: - Missing zero-input guard Correction: - Add explicit validation + test
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Go", "developer_needs": [ "governance", "tooling", "ci_integration", "evaluation_metrics" ], "moe_experts": [ "evaluation_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
8d2b3e4b4ffa9c07f7af81ee9bd82730b08cd978
train_21019
2026-01-01T00:00:00
Code-specialized model families and sizing tradeoffs
failure_analysis
advanced
Task: failure_analysis Topic: Code-specialized model families and sizing tradeoffs Difficulty: advanced Target language: Python Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Failure: - Initial patch broke edge case Reflection: - Missing zero-input guard Correction: - Add explicit validation + test
[]
{ "target_language": "Python", "developer_needs": [ "auditability", "reproducibility", "security_gates", "evaluation_metrics" ], "moe_experts": [ "coding_expert", "governance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
12d2b374ee8933a714724a399ed300c596958ee5
train_21020
2026-01-01T00:00:00
Extended context and repo-scale understanding
data_pipeline
expert
Task: data_pipeline Topic: Extended context and repo-scale understanding Difficulty: expert Target language: Bash Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Pipeline: Ingest → Normalize → Filter → Dedupe → Quality → Mix → Audit
[]
{ "target_language": "Bash", "developer_needs": [ "documentation", "tests_are_truth", "governance", "auditability" ], "moe_experts": [ "coding_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
d403b76750d8a1617547e54328780a693bf1a078
train_21021
2026-01-01T00:00:00
Multimodal dev workflows (docs, diagrams, traces)
patch_diff
intermediate
Task: patch_diff Topic: Multimodal dev workflows (docs, diagrams, traces) Difficulty: intermediate Target language: TypeScript Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Patch (diff-style): ```diff - if x == 0: - return 1/x + if x == 0: + raise ValueError('division by zero') ``` Acceptance: - Tests pass - No new regressions
[]
{ "target_language": "TypeScript", "developer_needs": [ "cost_latency_tradeoffs", "documentation", "evaluation_metrics", "auditability" ], "moe_experts": [ "data_curation_expert", "evaluation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
eaeba75462e8538e3cf6e585da44c827f8504e62
train_21022
2026-01-01T00:00:00
Agentic coding systems (plan→edit→test→reflect)
compare
intermediate
Task: compare Topic: Agentic coding systems (plan→edit→test→reflect) Difficulty: intermediate Target language: Java Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Compare: capability, cost, latency, reliability
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Java", "developer_needs": [ "evaluation_metrics", "auditability", "cost_latency_tradeoffs", "governance" ], "moe_experts": [ "security_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
6bf819ba8871a82b9c0ea5738d0323a1cdfebe6c
train_21023
2026-01-01T00:00:00
Code-specialized model families and sizing tradeoffs
review
advanced
Task: review Topic: Code-specialized model families and sizing tradeoffs Difficulty: advanced Target language: Bash Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[]
{ "target_language": "Bash", "developer_needs": [ "tests_are_truth", "governance", "repo_scale_reasoning", "cost_latency_tradeoffs" ], "moe_experts": [ "security_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
c0cee180d601c2951f056633f30a4f527efb9f84
train_21024
2026-01-01T00:00:00
SWE-bench style real-repo evaluation
failure_analysis
intermediate
Task: failure_analysis Topic: SWE-bench style real-repo evaluation Difficulty: intermediate Target language: JavaScript Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Failure: - Initial patch broke edge case Reflection: - Missing zero-input guard Correction: - Add explicit validation + test
[]
{ "target_language": "JavaScript", "developer_needs": [ "tests_are_truth", "security_gates", "reproducibility", "documentation" ], "moe_experts": [ "security_expert", "performance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
9aa68f3bb76746968f7a8b8dd792604463af561e
train_21025
2026-01-01T00:00:00
Extended context and repo-scale understanding
review
advanced
Task: review Topic: Extended context and repo-scale understanding Difficulty: advanced Target language: Bash Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Bash", "developer_needs": [ "security_gates", "ci_integration", "auditability", "tests_are_truth" ], "moe_experts": [ "coding_expert", "governance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
2bba3fc4ae467e6b048b1f7d7ca6a1f4380c7255
train_21026
2026-01-01T00:00:00
Self-improving agents and feedback loops
patch_diff
advanced
Task: patch_diff Topic: Self-improving agents and feedback loops Difficulty: advanced Target language: Python Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Patch (diff-style): ```diff - if x == 0: - return 1/x + if x == 0: + raise ValueError('division by zero') ``` Acceptance: - Tests pass - No new regressions
[]
{ "target_language": "Python", "developer_needs": [ "reproducibility", "cost_latency_tradeoffs", "documentation", "auditability" ], "moe_experts": [ "data_curation_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
08021b03d2768e03e3f4a67b9abdac50997e87cd
train_21027
2026-01-01T00:00:00
Mixture-of-Experts (MoE) for code
patch_diff
advanced
Task: patch_diff Topic: Mixture-of-Experts (MoE) for code Difficulty: advanced Target language: TypeScript Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Patch (diff-style): ```diff - if x == 0: - return 1/x + if x == 0: + raise ValueError('division by zero') ``` Acceptance: - Tests pass - No new regressions
[]
{ "target_language": "TypeScript", "developer_needs": [ "evaluation_metrics", "ci_integration", "tests_are_truth", "auditability" ], "moe_experts": [ "coding_expert", "performance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
8804f586605f7ce8ff06c20c3c68502783ca8bb1
train_21028
2026-01-01T00:00:00
Mixture-of-Experts (MoE) for code
failure_analysis
intermediate
Task: failure_analysis Topic: Mixture-of-Experts (MoE) for code Difficulty: intermediate Target language: TypeScript Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Failure: - Initial patch broke edge case Reflection: - Missing zero-input guard Correction: - Add explicit validation + test
[]
{ "target_language": "TypeScript", "developer_needs": [ "reproducibility", "repo_scale_reasoning", "tooling", "auditability" ], "moe_experts": [ "security_expert", "coding_expert" ], "governance": { "audit_required": true, "tests_required": true } }
4b7aae3697fb1f003af4be4514136f24840822f3
train_21029
2026-01-01T00:00:00
Mixture-of-Experts (MoE) for code
data_pipeline
intermediate
Task: data_pipeline Topic: Mixture-of-Experts (MoE) for code Difficulty: intermediate Target language: Python Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Pipeline: Ingest → Normalize → Filter → Dedupe → Quality → Mix → Audit
[]
{ "target_language": "Python", "developer_needs": [ "reproducibility", "tooling", "evaluation_metrics", "tests_are_truth" ], "moe_experts": [ "coding_expert", "evaluation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
f4d664150e4789794d493d3e7182321dffc1fcf7
train_21030
2026-01-01T00:00:00
Latency, cost, and reliability optimization
failure_analysis
advanced
Task: failure_analysis Topic: Latency, cost, and reliability optimization Difficulty: advanced Target language: TypeScript Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Failure: - Initial patch broke edge case Reflection: - Missing zero-input guard Correction: - Add explicit validation + test
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "TypeScript", "developer_needs": [ "reproducibility", "documentation", "security_gates", "repo_scale_reasoning" ], "moe_experts": [ "governance_expert", "coding_expert" ], "governance": { "audit_required": true, "tests_required": true } }
9a7185c9b7219fece76ac2e9a75fc487e718a13c
train_21031
2026-01-01T00:00:00
Multimodal dev workflows (docs, diagrams, traces)
explain
advanced
Task: explain Topic: Multimodal dev workflows (docs, diagrams, traces) Difficulty: advanced Target language: SQL Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "SQL", "developer_needs": [ "security_gates", "tests_are_truth", "ci_integration", "evaluation_metrics" ], "moe_experts": [ "security_expert", "coding_expert" ], "governance": { "audit_required": true, "tests_required": true } }
077762be642420f7683c458f68bc884d3ce084e1
train_21032
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 Target language: Java Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Java", "developer_needs": [ "ci_integration", "reproducibility", "tooling", "cost_latency_tradeoffs" ], "moe_experts": [ "coding_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
1f256f3e212f8f7bf9304f9266814799d690d177
train_21033
2026-01-01T00:00:00
SWE-bench style real-repo evaluation
compare
expert
Task: compare Topic: SWE-bench style real-repo evaluation Difficulty: expert Target language: SQL Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Compare: capability, cost, latency, reliability
[]
{ "target_language": "SQL", "developer_needs": [ "cost_latency_tradeoffs", "tests_are_truth", "documentation", "security_gates" ], "moe_experts": [ "security_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
e4888a1ade2272b1a531f97cb9d1a325fa55cdda
train_21034
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 Target language: Rust Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Rust", "developer_needs": [ "evaluation_metrics", "documentation", "reproducibility", "auditability" ], "moe_experts": [ "evaluation_expert", "performance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
92a5bc9f74e9370469a3ecc24ae3a459736e5791
train_21035
2026-01-01T00:00:00
Governance, provenance, and licensing for code data
patch_diff
expert
Task: patch_diff Topic: Governance, provenance, and licensing for code data Difficulty: expert Target language: Python Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Patch (diff-style): ```diff - if x == 0: - return 1/x + if x == 0: + raise ValueError('division by zero') ``` Acceptance: - Tests pass - No new regressions
[]
{ "target_language": "Python", "developer_needs": [ "cost_latency_tradeoffs", "documentation", "ci_integration", "evaluation_metrics" ], "moe_experts": [ "agentic_systems_expert", "governance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
2304dea39fc90a2d1f1019e48d3501fae3657a16
train_21036
2026-01-01T00:00:00
Secure code generation and policy gates
review
intermediate
Task: review Topic: Secure code generation and policy gates Difficulty: intermediate Target language: Java Context: Regulated environment requiring audit trails. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Java", "developer_needs": [ "repo_scale_reasoning", "tooling", "reproducibility", "cost_latency_tradeoffs" ], "moe_experts": [ "performance_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
92a77c14d72843a238787ba51c64eb059684f951
train_21037
2026-01-01T00:00:00
Self-improving agents and feedback loops
agent_loop
intermediate
Task: agent_loop Topic: Self-improving agents and feedback loops Difficulty: intermediate Target language: Go Context: Regulated environment requiring audit trails. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Loop: Plan → Edit → Test → Reflect → Human gate
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Go", "developer_needs": [ "governance", "security_gates", "evaluation_metrics", "reproducibility" ], "moe_experts": [ "governance_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
ae97ed8990e3e90248c611b9e4f7ab9c0fcebc88
train_21038
2026-01-01T00:00:00
SWE-bench style real-repo evaluation
design
expert
Task: design Topic: SWE-bench style real-repo evaluation Difficulty: expert Target language: Python Context: Regulated environment requiring audit trails. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "Python", "developer_needs": [ "ci_integration", "evaluation_metrics", "documentation", "cost_latency_tradeoffs" ], "moe_experts": [ "governance_expert", "performance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
7ac96022d3f2c105cc1f9911547b40b8f2c0a326
train_21039
2026-01-01T00:00:00
Tool calling, sandboxes, and CI integration
design
intermediate
Task: design Topic: Tool calling, sandboxes, and CI integration Difficulty: intermediate Target language: JavaScript Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "JavaScript", "developer_needs": [ "repo_scale_reasoning", "documentation", "auditability", "ci_integration" ], "moe_experts": [ "evaluation_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
4dda38b527bb3a65522a495a85a9139c33ac6044
train_21040
2026-01-01T00:00:00
Secure code generation and policy gates
patch_diff
advanced
Task: patch_diff Topic: Secure code generation and policy gates Difficulty: advanced Target language: Rust Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Patch (diff-style): ```diff - if x == 0: - return 1/x + if x == 0: + raise ValueError('division by zero') ``` Acceptance: - Tests pass - No new regressions
[]
{ "target_language": "Rust", "developer_needs": [ "tests_are_truth", "ci_integration", "governance", "evaluation_metrics" ], "moe_experts": [ "coding_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
0a0e89a9ea1abe78f64d2cfc542f921ae8d57bd9
train_21041
2026-01-01T00:00:00
Multimodal dev workflows (docs, diagrams, traces)
code
intermediate
Task: code Topic: Multimodal dev workflows (docs, diagrams, traces) Difficulty: intermediate Target language: TypeScript Context: Regulated environment requiring audit trails. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "TypeScript", "developer_needs": [ "tests_are_truth", "governance", "tooling", "repo_scale_reasoning" ], "moe_experts": [ "security_expert", "evaluation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
803a7238698e9f76b43cbdcfd6c83042e77e7ee4
train_21042
2026-01-01T00:00:00
Latency, cost, and reliability optimization
agent_loop
advanced
Task: agent_loop Topic: Latency, cost, and reliability optimization Difficulty: advanced Target language: Python Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Loop: Plan → Edit → Test → Reflect → Human gate
[]
{ "target_language": "Python", "developer_needs": [ "evaluation_metrics", "governance", "reproducibility", "ci_integration" ], "moe_experts": [ "security_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
caeeb6734682d10361dd67949fc6382c74373aa1
train_21043
2026-01-01T00:00:00
Dataset curation pipelines (filter, dedupe, quality)
design
intermediate
Task: design Topic: Dataset curation pipelines (filter, dedupe, quality) Difficulty: intermediate Target language: TypeScript Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "TypeScript", "developer_needs": [ "evaluation_metrics", "repo_scale_reasoning", "governance", "tests_are_truth" ], "moe_experts": [ "data_curation_expert", "performance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
d58c8e51ad7de6c24cf8f4e347e48b27a782bde8
train_21044
2026-01-01T00:00:00
Mixture-of-Experts (MoE) for code
eval
expert
Task: eval Topic: Mixture-of-Experts (MoE) for code Difficulty: expert Target language: Go Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Eval: pass@k, time-to-green, regressions, diff size
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Go", "developer_needs": [ "governance", "cost_latency_tradeoffs", "tests_are_truth", "auditability" ], "moe_experts": [ "coding_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
c8c8c5d9a08fc7215c842bf5647a5298247590aa
train_21045
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 Target language: TypeScript Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Failure: - Initial patch broke edge case Reflection: - Missing zero-input guard Correction: - Add explicit validation + test
[]
{ "target_language": "TypeScript", "developer_needs": [ "governance", "cost_latency_tradeoffs", "repo_scale_reasoning", "tooling" ], "moe_experts": [ "agentic_systems_expert", "coding_expert" ], "governance": { "audit_required": true, "tests_required": true } }
04d39d98bd060bdc6a799330bbae28dcf5093cdd
train_21046
2026-01-01T00:00:00
Code-specialized model families and sizing tradeoffs
eval
expert
Task: eval Topic: Code-specialized model families and sizing tradeoffs Difficulty: expert Target language: Rust Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Eval: pass@k, time-to-green, regressions, diff size
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Rust", "developer_needs": [ "ci_integration", "auditability", "repo_scale_reasoning", "reproducibility" ], "moe_experts": [ "agentic_systems_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
e8cd808037595dccdd39861869af3869edd0ed16
train_21047
2026-01-01T00:00:00
SWE-bench style real-repo evaluation
eval
advanced
Task: eval Topic: SWE-bench style real-repo evaluation Difficulty: advanced Target language: Bash Context: Regulated environment requiring audit trails. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Eval: pass@k, time-to-green, regressions, diff size
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Bash", "developer_needs": [ "repo_scale_reasoning", "security_gates", "documentation", "evaluation_metrics" ], "moe_experts": [ "performance_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
f692e7be9925b729e092530858ce3f841e948592
train_21048
2026-01-01T00:00:00
SWE-bench style real-repo evaluation
failure_analysis
intermediate
Task: failure_analysis Topic: SWE-bench style real-repo evaluation Difficulty: intermediate Target language: SQL Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Failure: - Initial patch broke edge case Reflection: - Missing zero-input guard Correction: - Add explicit validation + test
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "SQL", "developer_needs": [ "cost_latency_tradeoffs", "evaluation_metrics", "auditability", "reproducibility" ], "moe_experts": [ "evaluation_expert", "performance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
8fd85ffb8e5150dc761286ea234fc6667377c7cf
train_21049
2026-01-01T00:00:00
Governance, provenance, and licensing for code data
patch_diff
expert
Task: patch_diff Topic: Governance, provenance, and licensing for code data Difficulty: expert Target language: SQL Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Patch (diff-style): ```diff - if x == 0: - return 1/x + if x == 0: + raise ValueError('division by zero') ``` Acceptance: - Tests pass - No new regressions
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "SQL", "developer_needs": [ "cost_latency_tradeoffs", "documentation", "security_gates", "tooling" ], "moe_experts": [ "evaluation_expert", "governance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
b788162b3f1bfceb92c177943f6a67439824a2b6
train_21050
2026-01-01T00:00:00
Code-specialized model families and sizing tradeoffs
review
expert
Task: review Topic: Code-specialized model families and sizing tradeoffs Difficulty: expert Target language: SQL Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[]
{ "target_language": "SQL", "developer_needs": [ "ci_integration", "governance", "tooling", "reproducibility" ], "moe_experts": [ "security_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
c9b7361fcc9cb465a2eedaa66ab49a47836248c5
train_21051
2026-01-01T00:00:00
Dataset curation pipelines (filter, dedupe, quality)
design
advanced
Task: design Topic: Dataset curation pipelines (filter, dedupe, quality) Difficulty: advanced Target language: Java Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Java", "developer_needs": [ "reproducibility", "security_gates", "cost_latency_tradeoffs", "documentation" ], "moe_experts": [ "performance_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
b0cca1f995efcd6694e0a1ceb22c16e071e64e7f
train_21052
2026-01-01T00:00:00
Extended context and repo-scale understanding
patch_diff
advanced
Task: patch_diff Topic: Extended context and repo-scale understanding Difficulty: advanced Target language: SQL Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Patch (diff-style): ```diff - if x == 0: - return 1/x + if x == 0: + raise ValueError('division by zero') ``` Acceptance: - Tests pass - No new regressions
[]
{ "target_language": "SQL", "developer_needs": [ "repo_scale_reasoning", "cost_latency_tradeoffs", "security_gates", "evaluation_metrics" ], "moe_experts": [ "performance_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
b62f9cf18b5ffb2182f8559ed1ba35eed72637a6
train_21053
2026-01-01T00:00:00
Extended context and repo-scale understanding
review
advanced
Task: review Topic: Extended context and repo-scale understanding Difficulty: advanced Target language: Go Context: Regulated environment requiring audit trails. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Go", "developer_needs": [ "reproducibility", "security_gates", "repo_scale_reasoning", "tooling" ], "moe_experts": [ "performance_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
88c77271b0341c6d190f8f20015c7241c430654b
train_21054
2026-01-01T00:00:00
Secure code generation and policy gates
failure_analysis
expert
Task: failure_analysis Topic: Secure code generation and policy gates Difficulty: expert Target language: C# Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Failure: - Initial patch broke edge case Reflection: - Missing zero-input guard Correction: - Add explicit validation + test
[]
{ "target_language": "C#", "developer_needs": [ "security_gates", "tests_are_truth", "reproducibility", "ci_integration" ], "moe_experts": [ "evaluation_expert", "coding_expert" ], "governance": { "audit_required": true, "tests_required": true } }
47205dcd54583344d1401cea39ad019c6367a5a1
train_21055
2026-01-01T00:00:00
Mixture-of-Experts (MoE) for code
review
expert
Task: review Topic: Mixture-of-Experts (MoE) for code Difficulty: expert Target language: Rust Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[]
{ "target_language": "Rust", "developer_needs": [ "governance", "reproducibility", "ci_integration", "documentation" ], "moe_experts": [ "coding_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
f4348e2ace761aaa5df8acf931d24fd2639d928b
train_21056
2026-01-01T00:00:00
Code-specialized model families and sizing tradeoffs
patch_diff
advanced
Task: patch_diff Topic: Code-specialized model families and sizing tradeoffs Difficulty: advanced Target language: JavaScript Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Patch (diff-style): ```diff - if x == 0: - return 1/x + if x == 0: + raise ValueError('division by zero') ``` Acceptance: - Tests pass - No new regressions
[]
{ "target_language": "JavaScript", "developer_needs": [ "governance", "security_gates", "documentation", "auditability" ], "moe_experts": [ "coding_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
cfae7e09df4e92acf2f6c1b6e1a034ad7514fc25
train_21057
2026-01-01T00:00:00
SWE-bench style real-repo evaluation
agent_loop
expert
Task: agent_loop Topic: SWE-bench style real-repo evaluation Difficulty: expert Target language: Bash Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Loop: Plan → Edit → Test → Reflect → Human gate
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Bash", "developer_needs": [ "repo_scale_reasoning", "reproducibility", "tests_are_truth", "tooling" ], "moe_experts": [ "evaluation_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
f7fb38abfd82faac7f45c2132b9a5c333760c794
train_21058
2026-01-01T00:00:00
Self-improving agents and feedback loops
review
expert
Task: review Topic: Self-improving agents and feedback loops Difficulty: expert Target language: JavaScript Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "JavaScript", "developer_needs": [ "tests_are_truth", "tooling", "repo_scale_reasoning", "security_gates" ], "moe_experts": [ "coding_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
b76a148e23237b63d2e9d6088429c9a20b7068c8
train_21059
2026-01-01T00:00:00
SWE-bench style real-repo evaluation
review
intermediate
Task: review Topic: SWE-bench style real-repo evaluation Difficulty: intermediate Target language: Python Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[]
{ "target_language": "Python", "developer_needs": [ "cost_latency_tradeoffs", "reproducibility", "tests_are_truth", "security_gates" ], "moe_experts": [ "evaluation_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
f2b0a8cdf2b1d1e5c7eb04bcc0551c89602067f1
train_21060
2026-01-01T00:00:00
Agentic coding systems (plan→edit→test→reflect)
design
expert
Task: design Topic: Agentic coding systems (plan→edit→test→reflect) Difficulty: expert Target language: Rust Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "Rust", "developer_needs": [ "documentation", "governance", "reproducibility", "cost_latency_tradeoffs" ], "moe_experts": [ "data_curation_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
fc29b09b1d31d8c2a9d9572984245a648d8348f8
train_21061
2026-01-01T00:00:00
Agentic coding systems (plan→edit→test→reflect)
review
advanced
Task: review Topic: Agentic coding systems (plan→edit→test→reflect) Difficulty: advanced Target language: SQL Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[]
{ "target_language": "SQL", "developer_needs": [ "repo_scale_reasoning", "evaluation_metrics", "ci_integration", "auditability" ], "moe_experts": [ "data_curation_expert", "evaluation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
1904b71906484f8e69973c39baa98716b486fb3b
train_21062
2026-01-01T00:00:00
Self-improving agents and feedback loops
code
intermediate
Task: code Topic: Self-improving agents and feedback loops Difficulty: intermediate Target language: SQL Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "SQL", "developer_needs": [ "security_gates", "evaluation_metrics", "documentation", "tooling" ], "moe_experts": [ "performance_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
4241c9ec8490360fc8eb770c7121903f7e2cdb62
train_21063
2026-01-01T00:00:00
Dataset curation pipelines (filter, dedupe, quality)
agent_loop
advanced
Task: agent_loop Topic: Dataset curation pipelines (filter, dedupe, quality) Difficulty: advanced Target language: SQL Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Loop: Plan → Edit → Test → Reflect → Human gate
[]
{ "target_language": "SQL", "developer_needs": [ "documentation", "reproducibility", "repo_scale_reasoning", "cost_latency_tradeoffs" ], "moe_experts": [ "performance_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
f5bda5832f8a7529af123f2d95c16e326198571e
train_21064
2026-01-01T00:00:00
Multimodal dev workflows (docs, diagrams, traces)
design
expert
Task: design Topic: Multimodal dev workflows (docs, diagrams, traces) Difficulty: expert Target language: Go Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Go", "developer_needs": [ "ci_integration", "tooling", "documentation", "governance" ], "moe_experts": [ "coding_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
3a7691b7d9ee067e57a9abc78ecc306b236645ba
train_21065
2026-01-01T00:00:00
Tool calling, sandboxes, and CI integration
patch_diff
expert
Task: patch_diff Topic: Tool calling, sandboxes, and CI integration Difficulty: expert Target language: Bash Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Patch (diff-style): ```diff - if x == 0: - return 1/x + if x == 0: + raise ValueError('division by zero') ``` Acceptance: - Tests pass - No new regressions
[]
{ "target_language": "Bash", "developer_needs": [ "governance", "security_gates", "tests_are_truth", "repo_scale_reasoning" ], "moe_experts": [ "governance_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
7db75282fc9df12cd5db6a75e0ad9eef7310b73d
train_21066
2026-01-01T00:00:00
Latency, cost, and reliability optimization
compare
intermediate
Task: compare Topic: Latency, cost, and reliability optimization Difficulty: intermediate Target language: Python Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Compare: capability, cost, latency, reliability
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Python", "developer_needs": [ "repo_scale_reasoning", "auditability", "tests_are_truth", "cost_latency_tradeoffs" ], "moe_experts": [ "performance_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
4ed6169c9cdd118daf8da64d47d46cfa8b7f13d6
train_21067
2026-01-01T00:00:00
Self-improving agents and feedback loops
design
intermediate
Task: design Topic: Self-improving agents and feedback loops Difficulty: intermediate Target language: Go Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "Go", "developer_needs": [ "documentation", "repo_scale_reasoning", "security_gates", "reproducibility" ], "moe_experts": [ "governance_expert", "performance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
10fc9770090acc4e4c53c4993955290c28858a51
train_21068
2026-01-01T00:00:00
Governance, provenance, and licensing for code data
compare
expert
Task: compare Topic: Governance, provenance, and licensing for code data Difficulty: expert Target language: Python Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Compare: capability, cost, latency, reliability
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Python", "developer_needs": [ "governance", "evaluation_metrics", "repo_scale_reasoning", "documentation" ], "moe_experts": [ "performance_expert", "governance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
421f67df90eb0ebfaf66e9429038a959963d1d7d
train_21069
2026-01-01T00:00:00
Latency, cost, and reliability optimization
failure_analysis
expert
Task: failure_analysis Topic: Latency, cost, and reliability optimization Difficulty: expert Target language: JavaScript Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Failure: - Initial patch broke edge case Reflection: - Missing zero-input guard Correction: - Add explicit validation + test
[]
{ "target_language": "JavaScript", "developer_needs": [ "repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs", "ci_integration" ], "moe_experts": [ "coding_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
010eeee1b4195e0f9f09e072f90d08280db83a1f
train_21070
2026-01-01T00:00:00
Code-specialized model families and sizing tradeoffs
code
advanced
Task: code Topic: Code-specialized model families and sizing tradeoffs Difficulty: advanced Target language: C# Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "C#", "developer_needs": [ "evaluation_metrics", "security_gates", "cost_latency_tradeoffs", "reproducibility" ], "moe_experts": [ "performance_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
bef3ec228a81e0a139d3f6c00ff46377f1a3da85
train_21071
2026-01-01T00:00:00
Multimodal dev workflows (docs, diagrams, traces)
code
advanced
Task: code Topic: Multimodal dev workflows (docs, diagrams, traces) Difficulty: advanced Target language: SQL Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "SQL", "developer_needs": [ "governance", "reproducibility", "documentation", "security_gates" ], "moe_experts": [ "evaluation_expert", "performance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
6f7b2f8730c8ce252fc697ff5f3aea7e498a08ea
train_21072
2026-01-01T00:00:00
Self-improving agents and feedback loops
agent_loop
intermediate
Task: agent_loop Topic: Self-improving agents and feedback loops Difficulty: intermediate Target language: TypeScript Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Loop: Plan → Edit → Test → Reflect → Human gate
[]
{ "target_language": "TypeScript", "developer_needs": [ "auditability", "cost_latency_tradeoffs", "tests_are_truth", "reproducibility" ], "moe_experts": [ "coding_expert", "evaluation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
7958aa65de35f0d6ba2b616725f3dfb8cdb1d5ed
train_21073
2026-01-01T00:00:00
Extended context and repo-scale understanding
design
intermediate
Task: design Topic: Extended context and repo-scale understanding Difficulty: intermediate Target language: JavaScript Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "JavaScript", "developer_needs": [ "cost_latency_tradeoffs", "ci_integration", "documentation", "governance" ], "moe_experts": [ "security_expert", "governance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
117d5aa3bb53de119878b3ebc8efdf8727a4b2d4
train_21074
2026-01-01T00:00:00
Governance, provenance, and licensing for code data
explain
advanced
Task: explain Topic: Governance, provenance, and licensing for code data Difficulty: advanced Target language: Python Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "Python", "developer_needs": [ "security_gates", "tests_are_truth", "cost_latency_tradeoffs", "governance" ], "moe_experts": [ "performance_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
d26e3d4cd4d201e949d9396e476c26f611293264
train_21075
2026-01-01T00:00:00
SWE-bench style real-repo evaluation
explain
expert
Task: explain Topic: SWE-bench style real-repo evaluation Difficulty: expert Target language: Java Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "Java", "developer_needs": [ "tooling", "repo_scale_reasoning", "security_gates", "tests_are_truth" ], "moe_experts": [ "evaluation_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
0af844fc9d8fbcd96418ffa693816155b488f164
train_21076
2026-01-01T00:00:00
SWE-bench style real-repo evaluation
review
advanced
Task: review Topic: SWE-bench style real-repo evaluation Difficulty: advanced Target language: Java Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[]
{ "target_language": "Java", "developer_needs": [ "security_gates", "repo_scale_reasoning", "evaluation_metrics", "cost_latency_tradeoffs" ], "moe_experts": [ "data_curation_expert", "performance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
b49f4f98103efe58ef1d3d8e3471bf8b1c550e9f
train_21077
2026-01-01T00:00:00
Mixture-of-Experts (MoE) for code
code
expert
Task: code Topic: Mixture-of-Experts (MoE) for code Difficulty: expert Target language: Bash Context: Regulated environment requiring audit trails. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "Bash", "developer_needs": [ "governance", "documentation", "auditability", "tests_are_truth" ], "moe_experts": [ "performance_expert", "evaluation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
b9c5dca6bc639927cb35b65aaa14ed292641cd10
train_21078
2026-01-01T00:00:00
Model merging, distillation, and continued pretraining
eval
advanced
Task: eval Topic: Model merging, distillation, and continued pretraining Difficulty: advanced Target language: Rust Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Eval: pass@k, time-to-green, regressions, diff size
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Rust", "developer_needs": [ "tests_are_truth", "tooling", "cost_latency_tradeoffs", "security_gates" ], "moe_experts": [ "security_expert", "evaluation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
83a6e8c8ad16b50f012564f4ae0130fc4673433d
train_21079
2026-01-01T00:00:00
Model merging, distillation, and continued pretraining
agent_loop
intermediate
Task: agent_loop Topic: Model merging, distillation, and continued pretraining Difficulty: intermediate Target language: Java Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Loop: Plan → Edit → Test → Reflect → Human gate
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Java", "developer_needs": [ "repo_scale_reasoning", "evaluation_metrics", "tests_are_truth", "documentation" ], "moe_experts": [ "governance_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
59647fb77e1db8ebf504a56fea144566536fb4bc
train_21080
2026-01-01T00:00:00
Governance, provenance, and licensing for code data
code
intermediate
Task: code Topic: Governance, provenance, and licensing for code data Difficulty: intermediate Target language: SQL Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "SQL", "developer_needs": [ "cost_latency_tradeoffs", "evaluation_metrics", "repo_scale_reasoning", "documentation" ], "moe_experts": [ "data_curation_expert", "governance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
89f7fb12f2e22ce6a33c1af98e73b6939bce68b9
train_21081
2026-01-01T00:00:00
Reasoning-first coding models and tunable deliberation
failure_analysis
advanced
Task: failure_analysis Topic: Reasoning-first coding models and tunable deliberation Difficulty: advanced Target language: SQL Context: Regulated environment requiring audit trails. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Failure: - Initial patch broke edge case Reflection: - Missing zero-input guard Correction: - Add explicit validation + test
[]
{ "target_language": "SQL", "developer_needs": [ "ci_integration", "governance", "tests_are_truth", "tooling" ], "moe_experts": [ "evaluation_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
1e85b8008fb78fa3ba0bd6a5431bcc5b6027c08a
train_21082
2026-01-01T00:00:00
Governance, provenance, and licensing for code data
review
expert
Task: review Topic: Governance, provenance, and licensing for code data Difficulty: expert Target language: C# Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[]
{ "target_language": "C#", "developer_needs": [ "ci_integration", "documentation", "cost_latency_tradeoffs", "tests_are_truth" ], "moe_experts": [ "data_curation_expert", "evaluation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
ac2d0f906a277b80239bdacbe4248c6b3b80056e
train_21083
2026-01-01T00:00:00
Code-specialized model families and sizing tradeoffs
failure_analysis
intermediate
Task: failure_analysis Topic: Code-specialized model families and sizing tradeoffs Difficulty: intermediate Target language: JavaScript Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Failure: - Initial patch broke edge case Reflection: - Missing zero-input guard Correction: - Add explicit validation + test
[]
{ "target_language": "JavaScript", "developer_needs": [ "governance", "evaluation_metrics", "documentation", "security_gates" ], "moe_experts": [ "performance_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
80c9cd241f1e3e86653e5dd2f1599ecc7a47b1b2
train_21084
2026-01-01T00:00:00
Agentic coding systems (plan→edit→test→reflect)
design
advanced
Task: design Topic: Agentic coding systems (plan→edit→test→reflect) Difficulty: advanced Target language: Go Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "Go", "developer_needs": [ "documentation", "reproducibility", "evaluation_metrics", "governance" ], "moe_experts": [ "security_expert", "governance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
37bff38f43bfe58403adce4f8d2a6cda06b7c1e2
train_21085
2026-01-01T00:00:00
SWE-bench style real-repo evaluation
explain
expert
Task: explain Topic: SWE-bench style real-repo evaluation Difficulty: expert Target language: JavaScript Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "JavaScript", "developer_needs": [ "cost_latency_tradeoffs", "tooling", "security_gates", "ci_integration" ], "moe_experts": [ "agentic_systems_expert", "performance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
d22c282888083e14cc17e8499cc90d2b3137822e
train_21086
2026-01-01T00:00:00
Tool calling, sandboxes, and CI integration
failure_analysis
advanced
Task: failure_analysis Topic: Tool calling, sandboxes, and CI integration Difficulty: advanced Target language: Python Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Failure: - Initial patch broke edge case Reflection: - Missing zero-input guard Correction: - Add explicit validation + test
[]
{ "target_language": "Python", "developer_needs": [ "evaluation_metrics", "tests_are_truth", "auditability", "cost_latency_tradeoffs" ], "moe_experts": [ "evaluation_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
14d4813a8238321d56dbc44e2efff303f8192991
train_21087
2026-01-01T00:00:00
Secure code generation and policy gates
review
advanced
Task: review Topic: Secure code generation and policy gates Difficulty: advanced Target language: Java Context: Regulated environment requiring audit trails. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "Java", "developer_needs": [ "tooling", "documentation", "evaluation_metrics", "reproducibility" ], "moe_experts": [ "performance_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
199b600466492d492eeec002eec48ee17d10d342
train_21088
2026-01-01T00:00:00
Extended context and repo-scale understanding
code
advanced
Task: code Topic: Extended context and repo-scale understanding Difficulty: advanced Target language: Python Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Scaffold: ```python def agent_loop(plan, edit, test, issue, max_iters=4): history = [] p = plan(issue) for _ in range(max_iters): patch = edit(issue, p) ok, report = test(patch) history.append({"plan": p, "ok": ok}) if ok: return patch, history p = p + " | refine" return patch, history ```
[]
{ "target_language": "Python", "developer_needs": [ "governance", "auditability", "security_gates", "ci_integration" ], "moe_experts": [ "coding_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
aa89f9031ed37e4fd5ede39d21d4c4e5cb011a56
train_21089
2026-01-01T00:00:00
Latency, cost, and reliability optimization
explain
intermediate
Task: explain Topic: Latency, cost, and reliability optimization Difficulty: intermediate Target language: TypeScript Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[]
{ "target_language": "TypeScript", "developer_needs": [ "cost_latency_tradeoffs", "tooling", "governance", "evaluation_metrics" ], "moe_experts": [ "evaluation_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
e506a08a0c7ccfe55ab320de9c75d2fb2c25f333
train_21090
2026-01-01T00:00:00
Self-improving agents and feedback loops
agent_loop
intermediate
Task: agent_loop Topic: Self-improving agents and feedback loops Difficulty: intermediate Target language: TypeScript Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Loop: Plan → Edit → Test → Reflect → Human gate
[]
{ "target_language": "TypeScript", "developer_needs": [ "repo_scale_reasoning", "cost_latency_tradeoffs", "reproducibility", "governance" ], "moe_experts": [ "performance_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
7958aa65de35f0d6ba2b616725f3dfb8cdb1d5ed
train_21091
2026-01-01T00:00:00
Self-improving agents and feedback loops
data_pipeline
expert
Task: data_pipeline Topic: Self-improving agents and feedback loops Difficulty: expert Target language: Java Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Pipeline: Ingest → Normalize → Filter → Dedupe → Quality → Mix → Audit
[]
{ "target_language": "Java", "developer_needs": [ "auditability", "evaluation_metrics", "cost_latency_tradeoffs", "repo_scale_reasoning" ], "moe_experts": [ "security_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
16d363ff757c156e3f93487980f1de1b997f5e54
train_21092
2026-01-01T00:00:00
Multimodal dev workflows (docs, diagrams, traces)
patch_diff
expert
Task: patch_diff Topic: Multimodal dev workflows (docs, diagrams, traces) Difficulty: expert Target language: JavaScript Context: Regulated environment requiring audit trails. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Patch (diff-style): ```diff - if x == 0: - return 1/x + if x == 0: + raise ValueError('division by zero') ``` Acceptance: - Tests pass - No new regressions
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "JavaScript", "developer_needs": [ "cost_latency_tradeoffs", "tests_are_truth", "reproducibility", "security_gates" ], "moe_experts": [ "agentic_systems_expert", "coding_expert" ], "governance": { "audit_required": true, "tests_required": true } }
5a5c21dce1a2ed0d97bce98c3ea416adea9d3b78
train_21093
2026-01-01T00:00:00
Dataset curation pipelines (filter, dedupe, quality)
explain
advanced
Task: explain Topic: Dataset curation pipelines (filter, dedupe, quality) Difficulty: advanced Target language: SQL Context: Regulated environment requiring audit trails. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "SQL", "developer_needs": [ "evaluation_metrics", "auditability", "tests_are_truth", "ci_integration" ], "moe_experts": [ "agentic_systems_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
afa0caac479b01969d89d5a9a0049104999f06a5
train_21094
2026-01-01T00:00:00
Tool calling, sandboxes, and CI integration
explain
advanced
Task: explain Topic: Tool calling, sandboxes, and CI integration Difficulty: advanced Target language: TypeScript Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "TypeScript", "developer_needs": [ "auditability", "evaluation_metrics", "cost_latency_tradeoffs", "reproducibility" ], "moe_experts": [ "agentic_systems_expert", "evaluation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
1fe4259e9d32735b9e09a47a42897b2fa97e271c
train_21095
2026-01-01T00:00:00
SWE-bench style real-repo evaluation
data_pipeline
advanced
Task: data_pipeline Topic: SWE-bench style real-repo evaluation Difficulty: advanced Target language: JavaScript Context: Offline/local deployment with limited compute. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Pipeline: Ingest → Normalize → Filter → Dedupe → Quality → Mix → Audit
[]
{ "target_language": "JavaScript", "developer_needs": [ "cost_latency_tradeoffs", "governance", "documentation", "tooling" ], "moe_experts": [ "governance_expert", "coding_expert" ], "governance": { "audit_required": true, "tests_required": true } }
fbaf1991db51564d4147aa26fdf65fcd28e613be
train_21096
2026-01-01T00:00:00
Mixture-of-Experts (MoE) for code
review
advanced
Task: review Topic: Mixture-of-Experts (MoE) for code Difficulty: advanced Target language: Go Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Review: correctness, security, performance, governance
[]
{ "target_language": "Go", "developer_needs": [ "reproducibility", "cost_latency_tradeoffs", "auditability", "tests_are_truth" ], "moe_experts": [ "performance_expert", "data_curation_expert" ], "governance": { "audit_required": true, "tests_required": true } }
617d91b037191add1855ce9f42b1a746f1cdb2e1
train_21097
2026-01-01T00:00:00
SWE-bench style real-repo evaluation
design
intermediate
Task: design Topic: SWE-bench style real-repo evaluation Difficulty: intermediate Target language: C# Context: Research team validating claims against real repos. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Design with risks, metrics, acceptance criteria
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "C#", "developer_needs": [ "tests_are_truth", "evaluation_metrics", "tooling", "documentation" ], "moe_experts": [ "data_curation_expert", "agentic_systems_expert" ], "governance": { "audit_required": true, "tests_required": true } }
b630594deae0d9e4e500df3ec1d2568a4445cd2e
train_21098
2026-01-01T00:00:00
Multimodal dev workflows (docs, diagrams, traces)
eval
expert
Task: eval Topic: Multimodal dev workflows (docs, diagrams, traces) Difficulty: expert Target language: Bash Context: High-traffic service with latency SLOs. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Eval: pass@k, time-to-green, regressions, diff size
[]
{ "target_language": "Bash", "developer_needs": [ "tests_are_truth", "tooling", "documentation", "security_gates" ], "moe_experts": [ "coding_expert", "security_expert" ], "governance": { "audit_required": true, "tests_required": true } }
bfb0ae90860c5a0abc4c5d526bdac730214cd09b
train_21099
2026-01-01T00:00:00
SWE-bench style real-repo evaluation
data_pipeline
intermediate
Task: data_pipeline Topic: SWE-bench style real-repo evaluation Difficulty: intermediate Target language: TypeScript Context: Large monorepo with flaky tests and strict CI. Produce expert-level, production-ready artifacts.
Facts: - Modern AI coding prioritizes correctness, evaluation, and governance. - Agentic loops with test gates outperform single-pass generation. Pipeline: Ingest → Normalize → Filter → Dedupe → Quality → Mix → Audit
[ "Identify failing behavior via tests", "Minimize change surface", "Apply patch", "Run targeted + full suite", "Verify no regressions" ]
{ "target_language": "TypeScript", "developer_needs": [ "evaluation_metrics", "documentation", "cost_latency_tradeoffs", "tests_are_truth" ], "moe_experts": [ "agentic_systems_expert", "performance_expert" ], "governance": { "audit_required": true, "tests_required": true } }
006bb37b57654e404c47841a60080bee06d3c9f3