T1 FIX7 Spec-Artifact Focused Review And Correction Proposal - Readme First

Date 2026-06-08. Author T1 (production Agent, Agent Data). Mode READ-ONLY production; KB report writes only. Execution mode AUTHOR_MODE_ONLY. Live mutation NO.

What this package is

Focused adversarial review of Codex's newly published package codex-fix7-spec-artifact-publish-finalization-2026-06-07 (status FIX7_SPEC_ARTIFACTS_PUBLISHED_READY_FOR_T1_FOCUSED_REVIEW). It continues the prior T1 verdict DESIGN_NEEDS_CODEX_CORRECTION_BEFORE_IMPLEMENTATION (review package t1-fix7-implementation-spec-full-adversarial-review-2026-06-07).

This is NOT implementation, NOT design editing, NOT a permit step. T1 keeps the independent-reviewer role; Codex keeps design-owner authority for FIX7. Where a gap is found, T1 writes a precise correction proposal and does not patch Codex docs.

Boundary held (verified by this macro's own actions)

No production DB / role / grant / trigger / function / scheduler / UI / REAL_RUN / permit / ledger / QT001 apply mutated. No DB objects created. No SQL applied live. No manifest activated. No ownership/ACL change. No permit opened. Stage 2.6B not advanced. No Codex design doc edited. Only KB review reports + checkpoint written.

Inputs read in full

13 Codex FIX7 finalization artifacts (00..11 + checkpoint); prior T1 go/no-go (doc 17); Codex implementation-spec-completion checkpoint; governing law knowledge/dev/laws/prompt-muc-tieu-mo-for-claude-code.md (Chuẩn 100000x v1.3, incl. Design-Only Macro Mode, Survey-Then-Design Gate, Constitutional Guardrails: no_hardcode_absolute, no_fake_PASS, evidence-first, reversible-by-default); design index (consulted via search). Read path OPEN (not READ_PATH_BLOCKED).

Headline result

DESIGN_NEEDS_CODEX_CORRECTION_WITH_PROPOSALS. The package is a genuine, substantial advance: 4 of 8 corrections are VERIFIED (14 bypass vectors, Level-B operator packet, T1 self-audit dashboard, same-human quorum control), and the design remains genuinely PG-first/native and scale-safe. But 4 corrections remain PARTIAL and force an implementer to GUESS — which the law's no-hardcode/no-guess discipline forbids before approval:

• The "Normative full DDL artifact" (doc 02) is a contract-level spec, NOT byte-level CREATE TABLE DDL; it explicitly defers the authoritative DDL to an unsurfaced "local artifact" and leaves inter-manifest FK targets, the bootstrap code_catalog root, and the exactly-one-typed-operand column sets unspecified.
• Gate/vector thresholds (e.g. SCALE_SAFE 600000ms / 1073741824 bytes, "exact 3 runs") read as adapter literals rather than sealed manifest values (disguised-hardcode risk).
• Hash key-maps (doc 04) do not specify bytea→text encoding for embedded SHA-256 values, leaving canonical hash non-deterministic across implementers.
• Directus cutover (doc 06) is internally ambiguous on base-table vs presentation-view read preservation — the original feasibility risk is not closed.

9 correction proposals (7 blocking, 2 advisory) in doc 13. No hardcode FAIL, no PG-native FAIL, no scale FAIL — verdict narrows precisely to NEEDS_CORRECTION.

Document map

01 correction-matrix verification (SUPERTRACK A); 02 27-child-contract DDL (B); 03 14 readiness gates (C); 04 7 hash key-maps (D); 05 14 bypass vectors (E); 06 Directus cutover (F); 07 Level-B packet (G); 08 self-audit dashboard (H); 09 same-human control (I); 10 zero/disguised-hardcode scan (J); 11 PG-native-driven (K); 12 feasibility/scale (L); 13 consolidated correction package (M); 14 final verdict (N).

Next step

Codex addresses the doc-13 proposals (or accepts the explicit downscope option in CP-01), republishes, then a short T1 re-review can flip to DESIGN_READY_FOR_CODEX_FINAL_APPROVAL -> Codex final approval. No implementation, permit, Stage 2.6B, REAL_RUN, or QT001 apply until then. Readiness stays BLOCKED; scale stays NOT_SAFE-until-evidence.