COMMITS
June 6, 2026
T
fix(evals/contracts): correct eval_family for map-territory per VAL-CONTRACT-011
Travis Boudreaux committed
T
fix(m1-scorecard): resolve 3 blocking scrutiny issues
Travis Boudreaux committed
June 1, 2026
T
eval(coverage): confirm leads collapse at power; close 7 gaps; objective coverage 13→17
Travis Boudreaux committed
T
feat(evals): close coverage gaps — 6 authored objective sets + StrategyQA + run batches
Travis Boudreaux committed
T
docs(progress): Wave E advanced — branch pushed, PR #1 opened; merge-to-main is the gated step
Travis Boudreaux committed
T
docs(progress): Wave E — merge/push to main blocked by permission gate (needs user approval)
Travis Boudreaux committed
T
eval(wave-d): fermi rework re-measure (-5pp -> +7.5pp); Wave C validation; isolation log fix
Travis Boudreaux committed
T
eval(wave-c): record 6 powered verdicts; headline — debugging ELEVATEs do not replicate
Travis Boudreaux committed
T
docs(progress): log Waves A–D status and per-wave adversarial validation
Travis Boudreaux committed
T
eval(systems): restore Meadows leverage-points table; add Wave C sweep harness
Travis Boudreaux committed
T
test(evals): expand 17 pairwise behavioral sets from 3 to 25 problems each
Travis Boudreaux committed
T
feat(evals): add abstention + binary-decision runners and balanced datasets
Travis Boudreaux committed
T
chore: gitignore scratch evals/results/local/ runs
Travis Boudreaux committed
T
docs(analysis): refresh eval synthesis, backup record, dashboard, quality report
Travis Boudreaux committed
T
docs(repo): eval-informed metadata, README banner, agent guidelines
Travis Boudreaux committed
T
refactor(skills): apply audit follow-ups to five-whys-plus, ooda, model-router/selection
Travis Boudreaux committed
T
feat(scientific-method): ship hypothesis-differential debugging, retire v2 prototype
Travis Boudreaux committed
T
eval(debugging): close post-improvement re-measure of full debugging cluster
Travis Boudreaux committed
T
data(evals): post-improvement verification (improve->re-measure)
Travis Boudreaux committed
T
feat(skills): apply audit best-practices to all 39 skills
Travis Boudreaux committed
T
data(evals): paired-experiment + capability-ladder results
Travis Boudreaux committed
T
feat(evals): judge rubrics + statistical power analysis
Travis Boudreaux committed
T
feat(skills): scientific-method-v2 — agent-native rework (0 -> +5.3pp)
Travis Boudreaux committed
T
chore(planning): file-based plan, findings, and session log
Travis Boudreaux committed
T
docs(analysis): evaluation suite, audit, and elevate-or-kill synthesis
Travis Boudreaux committed
T
data(evals): run1 results + per-skill verdicts
Travis Boudreaux committed
T
feat(experiments): worktree orchestrator + paired/stack/capability runners
Travis Boudreaux committed
T
feat(reviews): adversarial review + lift-committee tooling
Travis Boudreaux committed
T
feat(evals): SQLite store + self-contained HTML dashboard
Travis Boudreaux committed
T
feat(evals): authored datasets + HuggingFace ingester/classifier
Travis Boudreaux committed