Apply the agreed dataset stats, evaluation settings, results, files and usage, and trace statistics format to all 17 releases. Include complete evaluated-cohort success distributions and supported pass@k plots. Standardize 19 trace-statistics companion READMEs, including distinct SETA thinking-on/off evaluations. Preserve material evidence caveats inline. No task, score, trace, or provenance data changes.
16 days ago
Name models by Hugging Face ID in trace_stats plot links
Use kept instead of retained in README prose. Order per-stage breakdowns as build/setup, no-op, reference solution, then rule-based checks. Preserve task counts, validation findings and schema field names.
23 days ago
Use consistent short dataset names with v1 and v2 folders
