KB-1C47
Codex FIX7 Refactor Blueprint Critical Review - Readme First
2 min read Revision 1
fix7codexcritical-reviewread-only2026-06-08
00 - Readme First
Date: 2026-06-08 Reviewer: Codex, FIX7 design owner Mode: READ-ONLY production and READ-ONLY blueprint review. No blueprint document was edited.
Final status
FIX7_REFACTOR_BLUEPRINT_CODEX_CRITICAL_REVIEW_FAIL_HARDCODE_OR_PG_NATIVE_GAP
The blueprint is not approved for implementation-authoring planning yet. The official FIX7 design remains approved, but the construction blueprint contains blocking contradictions and a disguised-hardcode/PG-native violation.
Blocking findings
- The binding S00 neutralization/cutover/freeze target is derived by
owner=directus+ QT001 name pattern. This is name-pattern authority, contradicting G-PGNATIVE and the blueprint's own statement that S00 inventory is diagnostic only. - G-NOLEGACY is required green before PKG-F although its ACL clause can only become green after PKG-F revokes EXECUTE. PKG-D also attempts the full guard while all legacy entrypoints are still PUBLIC-executable. This is a phase deadlock/false-green risk.
- PKG-F says replace every captured legacy entrypoint with a fail-closed stub, while S15/doc02 say stub only apply/writer entrypoints. Rollback only restores writer/gateway source, not every replaced body.
- S15 rollback restores complete PUBLIC EXECUTE while the new plane remains present. The claimed dormant/superseded state is not an explicit atomic rollback action/proof.
- Owner/ACL snapshot and guards omit column ACLs (
pg_attribute.attacl) and do not explicitly prove effective privileges through role membership. permitis used ambiguously for operator authorization although birth/admission permit must remain blocked.
No implementation, Stage 2.6B, permit, REAL_RUN, QT001 apply, manifest activation, repoint, ACL cutover, or production mutation is authorized.