fix(evals): harden H1 methods and add synthetic QA pack - #53

Merged

ernestprovo23 merged 14 commits into

mainfrom

feat/h1-pilot-pack

Jul 19, 2026

Commits on Jul 17, 2026

docs(evals): plan H1 method hardening and pilot
ernestprovo23
committed
94e3f2a
feat(evals): freeze paid study provenance and blocks
ernestprovo23
committed
6b1577a
feat(evals): harden paid grading and blinding
ernestprovo23
committed
8038e9d
feat(evals): enforce confirmatory analysis gates
ernestprovo23
committed
523d17f
fix(evals): label paid and confirmatory reports honestly
ernestprovo23
committed
e69d99f
feat(evals): add frozen 24-task elite pilot pack
ernestprovo23
committed
acedc8f
fix(evals): bind confirmatory gates to study evidence
ernestprovo23
committed
0ef30af
docs(evals): index the H1 pilot evidence pack
ernestprovo23
committed
bf07ea0
docs(evals): reclassify exposed pilot pack as synthetic QA
ernestprovo23
committed
730e246
fix(evals): enforce defensible confirmatory guardrails
ernestprovo23
committed
c4afbee
fix(evals): recompute confirmatory bounds from raw evidence
ernestprovo23
committed
5b03c18
fix(evals): make rotating grader kappa invariant
ernestprovo23
committed
5855299

Commits on Jul 18, 2026

Merge remote-tracking branch 'origin/main' into feat/h1-pilot-pack
ernestprovo23
committed
64c5043

Commits on Jul 19, 2026

fix(evals): enforce paid blinding boundary
ernestprovo23
committed
1b67e81