KB-6648

T1-Max FIX7 Refactor Blueprint Final Independent Review - Readme First

5 min read Revision 1
fix7architecturemax-reviewrefactor-blueprintreadme

00 - T1-Max FIX7 Refactor Blueprint Final Independent Review - Readme First

Date: 2026-06-08 Reviewer: T1-Max (final internal adversarial reviewer for Agent Data) Macro: PROGRAM_REVIEW_MAX_FIX7_REFACTOR_BLUEPRINT_FINAL_ADVERSARIAL_AUDIT_AND_DIRECT_REVISION Mode: READ-ONLY production. Direct-revision of blueprint KB docs allowed. No production mutation.

What this is

The final internal adversarial review + direct-revision pass over the FIX7 Existing-System Refactor Execution Blueprint (t1-fix7-existing-system-refactor-execution-blueprint-2026-06-08/), after the independent XHigh pass. The blueprint's embedded XHigh/Max self-review AND the separate independent XHigh pass were treated as evidence only, NOT as final authority. Live production evidence, the governing law, and document consistency decided each verdict. Worst-case-implementer lens: every ambiguous sentence is assumed to be misread; every unlisted legacy object is assumed to become a bypass; every "routed later" item is assumed unsafe unless blocked now.

Final status

FIX7_REFACTOR_BLUEPRINT_MAX_REVIEWED_AND_REVISED_READY_FOR_CODEX_CRITICAL_REVIEW

Not a blocker: no hardcode/PG-native FAIL, read path open, no invariant changed. 7 real defects were found and directly fixed in the blueprint docs; this report records what and why.

Live evidence gathered (read-only, DB directus, 2026-06-08)

  • All 46 legacy qt001_* functions + sp_dot_birth_qt001_apply have proacl=NULL = PUBLIC EXECUTE, and none is SECURITY DEFINER. The XHigh 4-function sample generalizes to the entire legacy function set: every one is directly callable by any role today. This is the load-bearing fact behind MB-01.
  • 0 triggers on qt001 tables; 0 triggers invoke a qt001 function. There is no hidden trigger-based bypass vector in the legacy QT001 layer.
  • Counts confirmed exact: 20 tables, 46 functions, 196 directus-owned views (the 196 = 183 v_qt001_* + 13 other qt001-named views). The blueprint's "196 views" is correct (an initial narrow-prefix count of 183 was a query artifact; re-verified by relkind+owner).
  • Birth-gateway family is 10 fn_birth_* functions (doc 01 listed 5), including one SECURITY DEFINER (fn_birth_onboarding_full_scan) and one owned by workflow_admin (fn_birth_gate); all are DO_NOT_TOUCH, none is a FIX7 control object.

Findings summary (7 findings; all fixed in-blueprint; 0 blockers)

id track severity one-line fixed in
MA-01 A integrity P2 post-XHigh top-line status still ..._READY_FOR_CODEX_CRITICAL_REVIEW (skipped the Max gate) doc 00, 12, checkpoint
MB-01 B completeness P1 legacy neutralization scoped to a sampled "apply/writer" subset, not the complete set (all 46 fns PUBLIC-executable live) doc 01, 02, 04, 05, 06, 07
MB-02 B completeness P3 birth-family inventory listed 5 of 10 live fn_birth_*; no trigger-vector evidence doc 01
MC-01 C feasibility P1 G-DOT-NOOVERWRITE led with "DOT bodies" (source-text) - risked source-text as final authority doc 06, 08
MD-01 D rollback (folded) rollback bypass/gateway-overwrite re-open + atomicity made explicit doc 04, 05
ME-01 E cutover P2 owner/ACL rollback snapshot not enumerated (concreteness gap before REVOKE) doc 05, 07
MG-01 G packages P2 no fresh Codex re-audit before the two governance-change packages PKG-F/PKG-G doc 07
MH-01 H guards P2 guard-quality rules missing (vacuous-pass, NULL-strict, source-text-as-authority) doc 06

(MD-01 is recorded under SUPERTRACK D as a clarification folded into the MB-01/ME-01 fixes, not a separate numbered defect; counted within the 7 above are MA-01, MB-01, MB-02, MC-01, ME-01, MG-01, MH-01.)

Document map

Doc Track
00 This readme
01 A - source & checkpoint integrity
02 B - legacy-entrypoint completeness
03 C - G-DOT-NOOVERWRITE feasibility & authority
04 D - rollback reopens-bypass review
05 E - owner/ACL cutover snapshot & rehearsal
06 F - construction-order implementability
07 G - package split
08 H - test / guard
09 I/J - hardcode / disguised-hardcode / PG-native
10 K - cross-layer
11 L - direct revisions applied
12 final verdict

Boundary

This is Max review only. Next step after PASS is Codex independent critical review. Implementation, Stage 2.6B, permit, REAL_RUN, QT001 apply, manifest activation, and owner/ACL cutover all remain BLOCKED. No production object was created, altered, owned, granted, revoked, or executed; the only writes were blueprint-doc revisions, this Max report, and the two checkpoints.

Back to Knowledge Hub knowledge/dev/reports/architecture/t1-max-fix7-refactor-blueprint-review-2026-06-08/00-readme-first.md