docs(science): Phase A closure report — validity program A1–A7

[pixelstrade-dev]

Closes the loop on Phase A of the validity program

docs/validity-program-phase-a.md is the closure record for Phase A (PRs #57–#63): a table mapping each criticism from the external audit ("scientific validity of the measurements: 2/10") to the merged answer, what it is now honest to claim, and — stated plainly — what still caps the score: no further code moves validity much from here; the four remaining gates are empirical (Run 002 with ≥ 3 judge families, corpus v1 per the A6 sizing, human annotation of PIGA intent spaces, external replication).

The report itself went through the orchestrator's fact-check (~45 claims verified against committed artifacts). Findings applied before this PR:

• construct-card provenance corrected (feat(science): construct registry + Evidence Card v3 design — Phase A1 (orchestrator-verified) #57 shipped 7 cards; the 8th, PIGA, came with feat(science): CAIMS-PIGA v0 — judge-as-classifier intent-disambiguation instrument (Phase A7) #63);
• the Run 002 owner action made operationally complete (BOTH OPENWEIGHT_API_KEY and OPENWEIGHT_BASE_URL as Secrets plus the exact model identifier — Run 001 recorded that judge as skipped for the missing env var);
• "published anchors" narrowed to the one actually-published anchor (Shrout–Fleiss 1979 → 0.29) plus hand-computed anchors;
• "a lucky silent guess still scores 0" corrected to "earns zero behavior credit" (the composite keeps the coverage channel, ~0–12);
• the adversarial-review sentence now states the reviews are session history and only their outcomes (committed correction notes) are repo-verifiable.

Docs-only change; no code paths touched.

🤖 Generated with Claude Code

https://claude.ai/code/session_01Y7wiMwFa2D4P8zD9RhCgsv

---

Generated by Claude Code