AI Council Prompt — Independent review of Codex PIDX v0.3.3 findings

Date: 2026-07-15 Scope: independent static review of Codex's two PIDX v0.3.3 reports. Review only; no SQL execution, build, DDL/DML, production mutation, or candidate design edit.

Objective

Determine which Codex findings are correct, overstated, understated, or based on a misunderstanding. Do not defer to Codex's verdict. Evaluate the source SQL and PostgreSQL 16 behavior directly.

Canonical Codex reports to review

1. knowledge/dev/laws-new/reports/codex/codex-pidx-v0.3.3-focused-static-review-2026-06-23.md
2. knowledge/dev/laws-new/reports/codex/codex-pidx-v0.3.3-required-patches-2026-06-23.md

Required source package

• knowledge/dev/laws-new/workflow-manage/design/pidx-ddl-candidate-v0.3.3.sql.md
• knowledge/dev/laws-new/workflow-manage/design/pidx-build-access-idempotency-contract-v0.3.3.md
• knowledge/dev/laws-new/workflow-manage/design/pidx-test-plan-v0.3.3.md
• knowledge/dev/laws-new/workflow-manage/reports/codex-pidx-v0.3.2-focused-static-review-2026-06-23.md
• knowledge/dev/laws-new/workflow-manage/reports/codex-pidx-v0.3.2-required-patches-2026-06-23.md

Questions each reviewer must answer

1. Does omission of static pg_sequence facets and identity ownership/dependency links permit a false PRESENT_MATCHING? If yes, what severity is justified?
2. Does Guard A's generic pg_depend predicate match expected PIDX indexes/constraints/identity machinery on a clean build? Show the exact dependency rows or authoritative PG16 basis. Is rollback actually blocked?
3. Does CREATE TABLE ... LIKE ... INCLUDING ALL copy GENERATED ALWAYS identity, and does the following plain INSERT INTO archive SELECT * FROM live fail without OVERRIDING SYSTEM VALUE? Is this a new HIGH regression or only a non-blocking recipe defect?
4. Is S11 materially unsafe because it accepts partial/invalid/not-ready unique indexes?
5. Is the delimiter-based content hash realistically collision-unsafe for this governed field domain, and what severity is warranted?
6. Did Codex miss any stronger counterevidence, existing contract clause, or PostgreSQL behavior that closes one of these findings?
7. Final independent verdict: AGREE, PARTIAL_AGREE, or DISAGREE for each Codex finding, with corrected severity and exact patch if needed.

Evidence rules

• Cite exact KB document path + section/line or exact SQL fragment.
• Use PostgreSQL 16 primary documentation or catalog facts; label inference explicitly.
• Separate executable SQL from comments/prose.
• Do not treat a claimed prior test as proof unless its real output is present.
• No generic “PASS/FAIL”; paste the evidence used.

Expected council outputs

• knowledge/current-state/reports/council-review-pidx-v0.3.3-codex-gpt-2026-07-15.md
• knowledge/current-state/reports/council-review-pidx-v0.3.3-codex-gemini-2026-07-15.md
• knowledge/current-state/reports/council-review-pidx-v0.3.3-codex-claude-code-2026-07-15.md
• synthesis after at least two independent reviews: knowledge/current-state/reports/council-synthesis-pidx-v0.3.3-codex-review-2026-07-15.md

Correction gate

Codex must not silently overwrite the original reports. After council synthesis, publish a correction matrix identifying CONFIRMED, SEVERITY_CHANGED, RETRACTED, or NEW_FINDING, then update the canonical reports with a dated corrigendum and preserve revision history.