KB-2332

FIX7 Final Authority-Seal — Codex Rejection Defect Re-Verification (under TKT v0.2, 2026-06-11)

4 min read Revision 1
tool-kiem-thufix7authority-sealcodex-rejectiondefect-reverificationtkt-v022026-06-11
<!-- DOC_STATUS: ACTIVE_NON_AUTHORITY -->

FIX7 Final Authority-Seal — Codex Rejection Defect Re-Verification (2026-06-11)

Host: T1. Authority: NON_AUTHORITY. Codex consulted: NO. Production mutated: NO. Method: every defect re-verified from the actual governed file, freshly materialized from KB and executed — neither Codex's original claim nor T1's later "patched" claim is assumed.

Source under test (byte-exact fresh reconstruction)

  • authority_seal_encoder.py (rev2) materialized from knowledge/dev/laws/tool-kiem-thu/packets/fix7-authority-closure-2026-06-10/.
  • Materialized sha256 = 13344f92cafcaf0d07dcb21700bdb642f38b89351702e08080eacb0e957144b8.
  • Cross-confirmed against TWO governed sources: the project pin, and the N-node model JSON field S3_authority_seal_encoder_spec.encoder_sha256 — both equal 13344f92…. ⇒ reconstruction is byte-exact, not behavioral-only.
  • Probe harness codex_probes.py (rev1) reconstructed byte-exact (sha 112b4ec5…).

Re-verification table (each row = real run, real exit)

# Codex-named defect probe (mutated input) expected actual PASS emitted? digest/seal emitted? verdict
1 invalid hash accepted N7 canonicalizer_sha256="NOT_A_SHA" (CP1) reject SEAL_FIELD_NOT_HEX NO NO CLOSED
2 empty id accepted N7 approval_event_id="" (CP2) reject SEAL_FIELD_EMPTY NO NO CLOSED
3 empty signer accepted N8 sealed_by="" (CP3) / N7 approver_identity="" (CP11) reject SEAL_FIELD_EMPTY NO NO CLOSED
4 wrong revision accepted P7 pinned_canonicalizer_revision="not-an-int" (CP5) / "-1" (SP2) reject SEAL_FIELD_BAD_INT NO NO CLOSED
5 wrong byte count accepted P7 pinned_canonicalizer_utf8_bytes="-1" (CP6) / "0" (CP13) reject SEAL_FIELD_BAD_INT NO NO CLOSED
6 empty report set accepted report_documents_digest([]) (CP7) reject SEAL_REPORT_SET_EMPTY NO NO CLOSED
7 duplicate report set accepted dup record (CP8) / dup doc-id diff rev (SP10) reject SEAL_REPORT_SET_DUPLICATE NO NO CLOSED
8 N6 active_corpus_sha256 was REHEARSAL into real seal encode_real_n7(provenance=REHEARSAL) (CP17) reject SEAL_PROVENANCE_REHEARSAL_BLOCKED NO NO CLOSED
9 report doc-id / revision malformed ("","1") (CP15) / ("doc","x") (CP16) reject SEAL_REPORT_DOC_ID_INVALID / ..._REVISION_INVALID NO NO CLOSED
10 byte-count↔hash independence (P7 verify) byte ok+hash wrong (SP3); hash ok+byte wrong (SP4) verify=False False / False NO NO CLOSED

Plus: candidate-class-as-authority (SP5 → SEAL_PROVENANCE_FORBIDDEN_CLASS), missing provenance (CP18 → SEAL_PROVENANCE_MISSING), forbidden class (CP19 → SEAL_PROVENANCE_FORBIDDEN_CLASS), valid-classes-but-no-real-N6 (SP6 → SEAL_REAL_N6_NOT_AVAILABLE), extra authority field (SP8 → SEAL_INPUT_EXTRA), prose-only pin (SP9 → SEAL_PROSE_ONLY_PIN_REJECTED).

Aggregate runs (captured exit codes)

  • python3 authority_seal_encoder.py --selftest48/48 PASS, exit 0.
  • python3 codex_probes.py19/19 REJECTED, 0 fail-open, exit 0.
  • python3 sp_align_probes.py (this lane's independent suite) → 10/10 fail-closed, exit 0.

Verdict

Every defect Codex named in CODEX_FIX7_FINAL_AUTHORITY_SEAL_REJECT is re-verified CLOSED on the current governed encoder, by direct bad-input execution — not by trusting prose. No patch was required in this lane (live proof showed no residual defect). The encoder is fail-closed. No contract defect remains.

The only thing the encoder still refuses to do — produce a real N7/N8/P7 — is by design (standing true blocker SEAL_REAL_N6_NOT_AVAILABLE + authority inputs), an authority/owner gap, not an engineering fail-open.

Back to Knowledge Hub knowledge/dev/reports/architecture/fix7-final-authority-seal-codex-rejection-defect-reverification-2026-06-11.md