KB-714B

FIX7 Refactor Blueprint - XHigh Adversarial Review

5 min read Revision 1
fix7architecturerefactor-blueprintxhigh-review

09 - T1-XHigh Adversarial Review

Reviewer posture: assume the High blueprint is wrong until each of the 15 checks proves otherwise. Findings that require a blueprint change are marked XHIGH_FINDING; the revision and re-check are recorded in doc 11.

# check verdict evidence / finding
1 confused build-from-scratch with refactor? PASS doc 00/02 explicitly model a parallel green-field qt001_cp plus repoint/freeze of live legacy; not a from-nothing build
2 missed any existing live object? PASS-with-note 20 tables/46 fns/196 views/birth gateway/DOTs inventoried live; iu_core/cutter_governance/sandbox_tac schemas hold no QT001 control objects; S00 re-baseline + G-UNKNOWN-ZERO catch residue
3 classified any object vaguely? PASS vocabulary used throughout; 196 views bulk-classed LEGACY_DEPRECATE as one homogeneous class (acceptable; S00 itemizes before freeze)
4 proposed adding something that already exists? PASS birth gateway referenced via #26 only; catalog root confirmed absent; no duplicate add
5 proposed modifying something that should be frozen? PASS DOT-118/119 + birth gateway are DO_NOT_TOUCH; legacy freeze gated behind proof
6 missed a legacy object that can override the new design? XHIGH_FINDING (XH-2) the new manifest path is proven non-reachable-to-legacy by #11, but the legacy apply/writer functions remain directus-owned and PUBLIC-executable between activation (S14) and owner/ACL cutover (S16). A direct caller of fn_qt001_plan_v5/legacy apply bypasses the repoint - this is exactly the FIX2/FIX3 "PUBLIC EXECUTE / writer public bypass" Codex rejected
7 introduced hardcode or disguised hardcode? PASS counts 27/11/14/7 are design invariants; birth anchor 1,210,928 is a reference baseline for G-BIRTH-NEUTRAL, not an authority literal; all thresholds resolve to sealed #05/#06 rows
8 violated PG-first/native/driven? PASS every guard reads PG catalog/data/recomputed hashes; G-NOLEGACY is structural #11 closure, not a name list
9 omitted rollback for any step? PASS doc 05 covers S01-S18; S00/S11/S19 are read-only
10 omitted test/guard for any step? XHIGH_FINDING (XH-3) S05 operator_operand_compatibility had no dedicated guard mapping (typed-operand coverage was implicit)
11 created a circular dependency? PASS FK cycles broken by deferred ALTERs (S08); package sequence is linear
12 missed Directus read-path impact? XHIGH_FINDING (XH-4) doc 04 S16 / doc 07 PKG-G said "REVOKE directus/PUBLIC authority" without scoping it to qt001_cp control objects; an implementer could over-revoke and break the live Directus CMS (directus_* system tables + business base tables Directus legitimately owns)
13 confused runtime-evidence with authority? PASS 11 runtime-evidence tables explicitly non-authority; G-RUNTIME-NONAUTH enforces count and exclusion from 27
14 altered the 27/11/14/7 invariants? PASS all additions are DATA rows or non-authority tables; exact-set seal verifies 27
15 allowed Stage 2.6B or apply path too early? PASS all apply/permit/REAL_RUN/2.6B blocked across S00-S19; PKG-E.. operator-gated

Minor refinement (advisory, not blocking)

  • XH-1: doc 02 mapped qt001_runtime_config solely under storage_class_manifest #05, but it also holds legacy driver batch/runaway config. Its driver-config role maps to sealed adapter behavior, not retention. Refine the mapping note (no count/severity change).

XHIGH_FINDINGs requiring revision

  • XH-2 (P1, real - historical bypass): legacy apply/writer entrypoints remain PUBLIC/directus executable in the activation->cutover window. Revision: (a) bundle legacy-entrypoint neutralization (REVOKE EXECUTE from PUBLIC/directus + fail-closed rollback stub) into S15/PKG-F, not deferred to PKG-G; (b) extend G-NOLEGACY to assert "legacy authoritative entrypoints executable by any non-owner role = 0", proving the legacy path is blocked, not merely unreachable from the new entrypoint.
  • XH-3 (minor): add guard G-OPERAND-TYPED (operand type matches operator_operand_compatibility for every rule/measurement-requirement operand) and map it to S05/S11.
  • XH-4 (P1, real - read-path safety): scope the ACL cutover to qt001_cp control objects + the enumerated legacy qt001_*/control objects only. Directus retains full authority over its own directus_* application tables and the business base tables it legitimately owns; only the control-plane authority and PUBLIC EXECUTE on control objects are revoked. Add G-DIRECTUS-APP-INTACT.

XHigh disposition

3 findings (XH-2, XH-3, XH-4) require blueprint revision; 1 advisory refinement (XH-1). All are revised in doc 11 and the affected checks (6, 10, 12) are re-run there. No invariant (27/11/14/7), no hard block, and no read-only/no-mutation posture is affected by the revisions.

Back to Knowledge Hub knowledge/dev/reports/architecture/t1-fix7-existing-system-refactor-execution-blueprint-2026-06-08/09-xhigh-adversarial-review.md