COMMITS
June 8, 2026
T
T
fix(m6-synthesis): resolve 5 blocking consistency issues (M6 scrutiny round-1)
Travis Boudreaux committed
June 7, 2026
T
T
T
T
fix(m5): replace stale scientific-method primary with fresh post-edit run (VAL-POWERED-014)
Travis Boudreaux committed
T
fix(m5-decisive-rerun): rerun 4 debugging skills on FROZEN 224-item decisive split, fix scorecard verdicts
Travis Boudreaux committed
T
feat(m5): add fresh scientific-method replication (+8.0pp, p=0.001) and scorecard updates
Travis Boudreaux committed
T
feat(m5): powered primary batch — 7 gated skills evaluated, all NO-LIFT
Travis Boudreaux committed
T
feat(m4): add trigger cards to 17 remaining contract-identified skills for full trigger-vs-full coverage
Travis Boudreaux committed
T
feat(m4): add trigger cards and strengthen redirects for 9 quarantine/kill candidates
Travis Boudreaux committed
T
feat(m4): rework six high-mechanism cohort skills from pre-registered specs
Travis Boudreaux committed
T
T
T
fix(m3-security-fpr-metric): recompute fpr_safe_items as combined FPR across all safe (answer=No) items
Travis Boudreaux committed
T
T
T
T
chore(m3-datasets): gitignore calibration scratch files (calib-batch/calib-sample)
Travis Boudreaux committed
T
June 6, 2026
T
T
fix(m3-datasets): fix fault-localization scorer parsing + truncation bugs, recalibrate debugging family
Travis Boudreaux committed
T
T
T
feat(m3-datasets): build candidate item pools per evaluation family with manifests
Travis Boudreaux committed
T
fix(m2-harness): remove unused imports droidJsonAsync/wilson from validate-judge.js
Travis Boudreaux committed
T
fix(m2-harness): calibration k>1 trials per item, distractor scoring on real dataset
Travis Boudreaux committed
T