research: rewrite-efficacy Study 4 — specificity-preservation constraint (H-4b), registration + harness - #719

Merged

devswha merged 10 commits into

devfrom

research/rewrite-efficacy-study4

Sep 3, 2026

Commits on Sep 2, 2026

research: register rewrite-efficacy Study 4 (specificity-preservation constraint, H-4b)

devswha
and
claude
committed
25da60f
research: Study 4 harness — bridge, runner, supervisor loop, analysis

devswha
and
claude
committed
a6e1f95
research(study4): dated note — gemini failed the bridge admission rule; stage 1 runs single-perceptual-judge

devswha
and
claude
committed
6ecf8a9
research(study4): amendment — bridge four API judge candidates under the registered admission rule

devswha
and
claude
committed
ed7be16
research(study4): admit judge-gemini-3.7-flash as the second perceptual judge (bridge AUC 1.000, rho 0.884)

devswha
and
claude
committed
055cb6f
research(study4): loop passes S4_JUDGES to the runner explicitly

devswha
and
claude
committed
b4fe9be
research(study4): keep claude's stdout message in rewrite failures for diagnosis

devswha
and
claude
committed
3fbaa63
research(study4): raise the rewrite timeout 600s -> 900s (execution note, Study 3 precedent)

devswha
and
claude
committed
23e105a
research(study4): reset-aware supervisor back-off and a larger pass budget

devswha
and
claude
committed
61e07d9

Commits on Sep 3, 2026

research(study4): stage 1 (ko) results — H-4b NOT supported; nothing ships

devswha
and
claude
committed
6ad0334