KB-7250 rev 2

FIX7 Codex N-number/N6 Targeted Repair Report

7 min read Revision 2
tool-kiem-thufix7codexn6tkt-v022026-06-11
<!-- DOC_STATUS: ACTIVE_NON_AUTHORITY -->

FIX7 — Codex N-number / N6 Targeted Repair Report (2026-06-11)

  • Host: T1 / Claude Code / MYTHOS. Codex consulted: NO. Owner approval requested: NO.
  • Final status: FIX7_CODEX_N_NUMBER_N6_REJECTION_TARGETED_REPAIR_READY.
  • Authority: NON_AUTHORITY engineering repair. Seals/ratifies/promotes nothing. No real N7/N8/P7. No production / PG / Directus / registry-row / system_issues mutation. T2 v0.2 framework NOT mutated. Tool-Kiem-Thu v0.2 NOT promoted.
  • Object IDs: TKT-OBJ-273..288 (collision-deconflicted: max canonical-committed = 262; a concurrent T2 NVSZ lane reserved 263..272 [PROPOSED/PENDING, V02-NVSZ-GOV-FOLD-1], so T1 takes 273..288; T1 182..200/208..216/225..253 + T2 201..207/217..224/254..262 intact).

0. What Codex rejected and what this macro repaired

Codex review CODEX_FIX7_N_NUMBER_OR_N6_REJECTED raised four FIX7-side blockers (a fifth, CODEX-T2-PACKET-INCOMPLETE, is T2-owned and was already repaired by T2; not touched here). All four FIX7-side blockers are now engineering-closed and replayed fail-closed on the published KB packets.

Codex blocker Exact surface State before Repair Proof Final verdict
CODEX-N6-DUP-DECLARED real_n6_provenance_verifier.py verify_corpus_membership() set conversion duplicate declared member accepted + certified reject duplicate declared ids before set conversion → N6_CORPUS_DUPLICATE, no cert replay P25 + selftest CLOSED
CODEX-N6-DUP-MANIFEST parse_hash_manifest() dict overwrite duplicate manifest path accepted + certified reject duplicate paths / malformed lines / non-64-lc-hex hashes before dict insert replay P26/P27/P28 + selftest CLOSED
CODEX-NNODE-ACTIVE-CONFLICT n7-approval-event-input-envelope.md §5 membership_sha256 (N1) stale (N1) label vs canonical (membership un-numbered) labels-only relabel → membership un-numbered (rev4) envelope-conformance 8/8 + n_node_checker 15/15 CLOSED
CODEX-ALIGNMENT-PACKET-INCOMPLETE alignment packet missing authority_seal_encoder.py + codex_probes.py 2 manifest-listed files absent at packet path publish byte-exact governed copies (== pins) fresh RERUN.sh PASS, G0–G4 CLOSED

1. Hash summary (before → after)

Artifact Before After
real_n6_provenance_verifier.py 277daf86…b67cf9cb 922e5246…80224385
n6_adversarial_probes.py a37d9b98…82e8a241 0f2c94d3…ce2865dfe (P25–P28 added)
real-N6 packet HASH_MANIFEST / packet_tree 43b4914a…752eea23 356a0cee…282a8b9b
real-N6 README / manifest.json / exit_codes.json / 2 logs (rev1) rev2 (counts 14→19, 27→31)
alignment authority_seal_encoder.py ABSENT at packet path published 13344f92…57144b8 (== pin)
alignment codex_probes.py ABSENT at packet path published 112b4ec5…465a3d2 (== pin)
alignment packet_tree 96d00b9e…03e83c1 96d00b9e…03e83c1 (unchanged — manifest already correct)
n7-approval-event-input-envelope.md rev3 (had (N1)) rev4 182a6d9f…b09f3a5fe1
n7-approval-event-input-envelope.json rev3 rev4 138a356b…d127cb55db0bd0187
N6 value / certificate d777e87c… / acf8d259… unchanged (engineering candidate; cert byte-identical)

2. End-to-end proofs (all re-run live; none trusted from prose)

2a. Real-N6 packet — fresh-from-KB clean-room bash RERUN.sh

  • cwd: fresh /tmp dir, all 18 files re-fetched from governed KB.
  • RECONSTRUCTION: OK (32 recheck-9 files + governed encoder 13344f92…).
  • selftest 19/19 PASS (exit 0); REAL-N6 VERIFY: OK n6=d777e87c… class=ENGINEERING_VERIFIED_CANDIDATE authority=NOT_A_SEAL; regenerated cert byte-identical to pinned acf8d259…; adversarial probes 31/31 fail-closed (any PASS emitted: False | any seal/official digest emitted: False); authority firewall 8/8; cert binding intact.
  • OVERALL: PASS (5/5 gates)RERUN_RESULT: PASS, exit 0.

2b. Alignment packet — fresh-from-KB clean-room bash RERUN.sh

  • step1 shasum -c HASH_MANIFEST.txt 5/5 OK (both previously-absent files now present); step2 packet_tree OK 96d00b9e…; step3 ledger G0_encoder_frozen / G1_encoder_selftest / G2_codex_probes / G3_n_node_checker / G4_supplemental all OK.
  • OVERALL_RESULT: PASSRERUN_RESULT: PASS, exit 0.

2c. Codex-owned duplicate replay (reproduced from rejection)

On the unpatched governed verifier 277daf86: DUP-DECLARED ACCEPTED+cert, DUP-MANIFEST ACCEPTED+cert, MALFORMED-MANIFEST ACCEPTED+cert → 3/3 FAIL-OPEN reproduced. On the patched verifier 922e5246: N6_CORPUS_DUPLICATE / N6_MANIFEST_DUPLICATE / N6_MANIFEST_MALFORMED_LINE, cert_emitted=False → 3/3 FAIL-CLOSED. See fix7-codex-rejection-probe-replay-report-2026-06-11.md.

2d. N-number envelope conformance

envelope_conformance_probe.py binds n_node_checker.py to the live rev4 envelope: 8/8 PASS — stale membership_sha256 (N1) absent, membership annotated un-numbered, checker canonical N1 = normalized_active_content_sha256[d], R4 rejects membership-value-under-N1, N2/N3/N6 labels still canonical.

3. Final reality checks

Question Answer
Duplicate declared corpus accepted? NO
Duplicate manifest record accepted? NO
Duplicate HASH_MANIFEST path accepted? NO
Certificate emitted for invalid duplicate? NO
N7 envelope stale N1 membership label remains? NO
Alignment packet load-bearing file missing? NO
Fresh reconstruction passes (both packets)? YES
Report-vs-file audit passes? YES
Authority firewall passes? YES (8/8)
N6 claimed Codex-ratified? NO (engineering candidate only)
N7/N8/P7 authored? NO / NO / NO
Production mutation? NO
T2 v0.2 framework mutated? NO

4. Remaining blockers (authority/owner only — NOT engineering)

Blocker Kind Actor Blocks next Codex review?
N-number table ratification (G-DOC-1) authority/doc owner/Codex No — engineering label reconciled; binding numbering needs ratification
SEAL_REAL_N6_NOT_AVAILABLE (G-AUTH-1) authority/provenance owner/operator + Codex gates real N6→seal; N6 stays candidate
A1/A2/A3/A5 + N8 signer + P7 ids (G-AUTH-2) authority owner/Codex real N7/N8/P7 inputs
OWN-1 standing do-not-approve (G-OWNER-1) owner owner seal authorization

All four are authority-only; no safe same-scope engineering work remains. Minimal next step: route a fresh independent Codex review of the patched real-N6 packet (356a0cee…), alignment packet (96d00b9e…), and N7 envelope rev4 — the four FIX7 engineering blockers are closed; only ratification/authority inputs remain.

Back to Knowledge Hub knowledge/dev/reports/architecture/fix7-codex-n-number-n6-targeted-repair-report-2026-06-11.md