01 - Codex Recheck-7 Failure Matrix

Source: codex-fix7-blueprint-recheck-7-byte-exact-seal-2026-06-09/ (00, 01, 03–08, 10, 11) + checkpoint-codex-fix7-blueprint-recheck-7-byte-exact-seal-2026-06-09.md.

Codex status: FIX7_REFACTOR_BLUEPRINT_CODEX_RECHECK_7_NEEDS_T1_FIX.

Accepted

Codex check	verdict
Active extractor	ACTIVE_EXTRACTOR_ACCEPTED
Seal hash graph (content-hash DAG)	SEAL_HASH_GRAPH_DAG_ACCEPTED
Invariants / boundaries	ACCEPTED
Hardcode / PG-native (runtime)	no new regression; prior acceptance intact
Membership	f2bda8…fe251 (independent prior evidence valid)

Blocked → finite correction list (doc 10) and the T1 fix

Codex check	verdict	root cause (verbatim sense)	T1 fix
Constitution-14	CONSTITUTION_14_CHECK_NEEDS_FIX	(1) anchor needs the checkpoint's own final revision recorded+read-back — only exists after write → can't complete finitely; (2) canonicalizer described but not pinned as one executable/versioned/hashed SSOT with invocation + test vectors; (3) multiple load-bearing descriptions can drift — Điều 14 requires one authority of one nature, all other copies explicit non-authority	doc 02 (anchor) + doc 03 (SSOT) + doc 04 (no-duplicate)
Anchor model	CODEX_ANCHOR_MODEL_NEEDS_FIX	"requires the Codex checkpoint's own platform revision recorded in/with its detached seal then read back for equality… writing that revision creates a later revision… revision-layer self-reference even though the content-hash DAG is acyclic" → choose a non-circular option + finite write/read sequence	doc 02 (Option 1: own-revision removed from authority; revision diagnostic; finite ordered seal)
Record encoding	RECORD_ENCODING_NEEDS_FIX	canonical doc identity not enforced (path aliases → duplicate logical records); marker literal/kind not one consistent pair; no executable canonicalizer pinned by path/version/hash/invocation/test-vector	docs 03 + 05 + 06
Codex sealing	CODEX_SEALING_NEEDS_T1_FIX	could not author a final seal: anchor not finitely executable + canonicalizer not uniquely pinned	resolved by A–E; Codex seals at recheck 8
Hardcode/PG-native (final gate)	HARDCODE_FINAL_NEEDS_FIX / PG_NATIVE_FINAL_NEEDS_FIX	"authority construction still permits disguised fixed behavior through an unpinned canonicalizer and duplicate/drifting load-bearing contract descriptions; a PG-driven design is not complete when the construction authority can be interpreted differently by different agents" → add G-CANONICAL-CONTRACT-SSOT-NO-DUPLICATE	doc 04 + doc 08
Authoring-planning	AUTHORING_PLANNING_NEEDS_T1_FIX	the 7-item finite correction list	all patched (doc 10)

The two decisive root causes

1. Revision-layer self-reference survived recheck-6: even with the content-hash DAG acyclic, making the recheck-N checkpoint's own platform revision load-bearing (recorded + read-back equality) is circular, because the revision exists only after the write that records it. → Blocker A.
2. No single executable SSOT + duplicate authority: the canonicalizer was prose, and the same load-bearing rules appeared in doc 00, the guards, and the reports, which can drift — an Article-14 violation. → Blockers B + C.

Both are closed in this pass: an executable canonicalizer SSOT (doc 03), all other text demoted to NON_AUTHORITY_EXPLANATION (doc 04), and the revision removed from authority (doc 02) — computed in doc 07.