49,000 training examples and 1,000 validation examples comprise this expert dataset for AI code generation. Developed by Within Us AI, it features supervision patterns based on test outcomes and agentic loops. The dataset was last updated on January 2, 2026.
Use Cases
- Training code generation models based on diff-first patching supervision.
- Developing agentic loops (plan→edit→test→reflect) for automated software engineering.
- Implementing governance and audit systems based on policy-gate awareness flags.
- Supervising tool-call traces for AI agents interacting with programming environments.
Strengths
- 49,000 training examples provide a substantial base for model development.
- Includes expert supervision patterns like tests-as-truth and diff-first patching.
- Features governance and audit flags for policy-aware applications.
Limitations
- Column-level documentation is absent; field semantics must be inferred after download.
- Row count is unknown for the total dataset, which may limit suitability assessment.
- Description metadata is limited; actual data quality requires manual inspection after download.
Provenance
- Source
- WithinUsAI
- Collection Method
- Likely contains expert-curated code examples with failure→reflection→correction loops.
- Freshness
- Last updated 2026-01-02 04:26:42; freshness should be verified.