fix(m5): replace stale scientific-method primary with fresh post-edit run (VAL-POWERED-014)
Run fresh post-edit PRIMARY for scientific-method on the ORIGINAL frozen 150-item SWE-bench set (EVAL_RUN=m5-primary-sci, claude-sonnet-4-6, CONC=4, isolation ON). The old run1 primary (+9.3pp, mtime Jun 1 03:51) predated the SKILL.md procedural rewrite (Jun 1 06:05, commit 4a176d9) and was replaced. Fresh primary: 93% vs 87% placebo, +5.4pp, p=0.027 (n=149), post-edit. Replication (m5-repl): 90% vs 82%, +8.0pp, p=0.001 (n=150), post-edit. Both same positive direction, both p<0.05 — ELEVATE confirmed on TWO genuinely post-edit samples. Updated both scorecard artifacts (JSON + MD) to point primary to the new result. VAL-POWERED-014 satisfied: primary mtime > SKILL.md mtime.
T
Travis Boudreaux committed
49d000e45878ec71d8cb337ac50c9a1c21be4ff9
Parent: fcb8c41