FIX7 Final Authority-Seal — Fail-Open + Provenance Patch — Master Report

• Date: 2026-06-11 Asia/Ho_Chi_Minh · Host: T1 / Claude Code / Mythos
• Lane: FIX7_FINAL_AUTHORITY_SEAL_FAILOPEN_AND_PROVENANCE_PATCH_MACRO_2026_06_11
• Final status: FIX7_FINAL_AUTHORITY_SEAL_SELF_CODEX_READY_AFTER_FAILOPEN_PATCH
• Standing authority blocker (Status-B condition, prominently surfaced): FINAL-AS-N6-PROVENANCE = SEAL_REAL_N6_NOT_AVAILABLE — no real non-rehearsal N1..N6 chain exists in this lane; creating/sealing it is a Codex/owner authority act, not done here.
• Production mutation: NO · Codex consulted: NO · Real N7/N8/P7 authored: NO / NO / NO · Real seal claimed: NO

0. What Codex rejected and what this lane did

Codex returned CODEX_FIX7_FINAL_AUTHORITY_SEAL_REJECT on two independent grounds, proved by 8 direct adversarial probes:

1. Fail-open encoder — NOT_A_SHA, empty IDs/signers, not-an-int revisions, -1 byte counts, empty + duplicate report sets all produced a 64-hex digest.
2. Provenance-blind N7 — the proposed N6 active_corpus_sha256 is classified REHEARSAL, yet nothing stopped it laundering into a real N7.
3. Governed-KB evidence missing — 7 required rehearsal files returned 404 (prose-only embedding instead of the actual files).

This lane reproduced the 8 probes as ACCEPTED on the pre-fix encoder, patched the encoder/spec/harnesses to fail closed, added a provenance class system, published the 7 governed-KB evidence files, extended the red-team to include every Codex probe class, and ran a full fresh self-Codex dry-run. It does not author any real seal and does not implement FIX7.

1. Before/after of Codex's 8 direct probes (full table in the probes report)

#	Probe	Before (pre-fix)	After (patched)
1	N7 canonicalizer_sha256="NOT_A_SHA"	ACCEPTED a38c6b2b…	REJECTED SEAL_FIELD_NOT_HEX
2	N7 approval_event_id=""	ACCEPTED dc2edfb4…	REJECTED SEAL_FIELD_EMPTY
3	N8 sealed_by=""	ACCEPTED f0d22aba…	REJECTED SEAL_FIELD_EMPTY
4	N8 report_documents_digest="NOT_A_SHA"	ACCEPTED d12bd027…	REJECTED SEAL_FIELD_NOT_HEX
5	P7 pinned_canonicalizer_revision="not-an-int"	ACCEPTED 113481ef…	REJECTED SEAL_FIELD_BAD_INT
6	P7 pinned_canonicalizer_utf8_bytes="-1"	ACCEPTED 079af996…	REJECTED SEAL_FIELD_BAD_INT
7	report_documents_digest([])	ACCEPTED a1b0e446…	REJECTED SEAL_REPORT_SET_EMPTY
8	duplicate report-document records	ACCEPTED 256351be…	REJECTED SEAL_REPORT_SET_DUPLICATE

Pre-fix: CODEX-PROBES: 0/19 REJECTED (accepted/fail-open=19). Post-fix: 19/19 REJECTED (accepted/fail-open=0). Notably the pre-fix provenance probes (CP17/18/19) produced the published fixture N7 digest 6225f265…, demonstrating the laundering risk concretely.

2. Blocker closure ledger (Codex final-seal blockers)

Codex blocker	Required correction	Status	Evidence
FINAL-AS-VALUE-GRAMMAR	per-field non-empty/type/grammar/range; reject bad hashes/IDs/revisions/byte-counts/signers/timestamps	CLOSED	value-grammar report; selftest 48/48; probes CP1–CP6/CP9–CP14 REJECTED
FINAL-AS-REPORT-SET	reject empty/duplicate/invalid report sets; deterministic sort	CLOSED	report-set report; probes CP7/CP8/CP15/CP16 REJECTED
FINAL-AS-PROVENANCE	provenance classes; reject REHEARSAL/missing/forbidden into real N7	CLOSED (engineering)	provenance report; probes CP17/CP18/CP19 REJECTED
FINAL-AS-KB-PACKET	publish actual governed rehearsal files; prove manifest/tree	CLOSED	governed-KB evidence report; 7 files now present (404→present)
FINAL-AS-N6-PROVENANCE (surfaced by the gate)	supply a real non-rehearsal ENGINEERING_VERIFIED_CANDIDATE N1..N6 chain, then seal	TRUE BLOCKER (owner/operator + Codex)	SEAL_REAL_N6_NOT_AVAILABLE; not faked
IMPL-OWNER-AUTHORIZATION	separate implementation macro after seal	OPEN (owner)	precondition checklist

3. Self-Codex dry-run (fresh, sequential, exit 0) — bash rehearsal/commands.sh → rc 0

Step	Result	Exit
authority_seal_encoder.py --selftest	48/48 PASS	0
authority_seal_rehearsal.py (N7→N8→P7, deterministic, rehearsal→real BLOCKED)	REHEARSAL OK	0
authority_seal_redteam.py (sequential, after rehearsal)	39/39 caught (incl. all Codex probe classes)	0
authority_seal_drift_check.py .	41/41 agree, drift 0	0
authority_seal_antihardcode.py	13/13 PASS (incl. provenance-laundering T6)	0
codex_probes.py (direct probes)	19/19 REJECTED (fail-closed)	0

Artifacts: packet_tree.sha256 = ac3f56f917f760760a71000b0c7a43c65cc40a5ceb0c468dd947dd6a579477dc (reproducible — identical across two consecutive full runs). Encoder sha256 = 13344f92cafcaf0d07dcb21700bdb642f38b89351702e08080eacb0e957144b8. cwd /tmp/fix7-failopen/packet.

4. Pins preserved (no engineering contradiction)

Fixture digests are byte-identical to the reviewed values after the patch: N7 6225f265…459bfd, N8 b1f001b6…75aa73, P7 3599f663…a7d541. Engineering pins unchanged: canonicalizer rev3 49c386a9…b734d0 (rev 3, 38756 B), Packet V3 tree b95df0a5…ca6d, membership f2bda8…fe251. The patch adds validation that runs before encoding; valid fixtures encode identically ⇒ no Packet V3 engineering contradiction.

5. Governed-KB evidence (FINAL-AS-KB-PACKET)

The 7 files Codex flagged 404 are now governed and re-fetched present (each content_length > 0): rehearsal/commands.sh, HASH_MANIFEST.txt, packet_tree.sha256, rehearsal-summary.json, exit_codes.json, stdout.log, stderr.log. Reconstruction model: fetch the byte-exact source files (encoder verified 13344f92…, round-trip content_length=35135 match) → run commands.sh → it regenerates the artifacts + manifest + tree deterministically (ac3f56f9…). See the governed-KB evidence report.

6. Honest scope / non-overclaim

• No real N7/N8/P7 was authored; the only digests produced are FIXTURE/NOT-A-SEAL.
• The provenance gate is engineering-complete, but the real N6 chain is not materialized or sealed here — that is SEAL_REAL_N6_NOT_AVAILABLE, a genuine owner/operator + Codex authority action. It was not faked.
• KB-stored generated artifacts are regenerated by commands.sh; the byte-exact reproducible inputs are the source files (whose hashes are listed in HASH_MANIFEST.txt). A fresh KB-fetch-and-rerun is the Codex-side verification step.
• No production / PG / Directus / registry / system_issues mutation; no REAL_RUN / QT001 / permit / activation / repoint / cutover; no registries-pivot; no auto-birth repair.

7. Deliverables (this lane)

Reports: master (this), codex-probes-before-after, value-grammar-validation, provenance-validation, report-set-validation, governed-kb-evidence-packet, redteam-expanded-codex-probes, self-codex-dry-run-after-patch. Updated files: encoder/spec.md/spec.json/redteam/drift/antihardcode/rehearsal + codex_probes.py + n7(md/json)/n8/p7 + checklist + 7 rehearsal evidence files + artifacts. Ledger rev7; checkpoint; current-state; object registry + 00-index governance.

8. Minimal safe next step

Route the patched closure packet to Codex for a fresh final-seal review (Codex re-runs §3 from governed KB; all probes now fail closed). The seal itself remains blocked on SEAL_REAL_N6_NOT_AVAILABLE (owner/operator + Codex must supply and seal a real ENGINEERING_VERIFIED_CANDIDATE N6 chain) and on owner implementation authorization. No implementation until both close.