OIC semantic code-start gate closure candidate

[veraxis-protocol]

Scope

Owner-authorized non-semantic closure of OD-001 through OD-004 from current main fc96d6f590a6f8ed12ecbb256f33f6de16d0c619.

PR #35 at 8fc72aba7c7de80e64de83a8f9f65184b002b9ac remains the historical audit/readiness evidence basis. This is a fresh closure branch; it does not implement semantics.

Owner decisions represented

• OD-001: CA-3 plus three deterministic, explicitly SYNTHETIC and FICTIONAL companion sources. Each byte set has an immutable SHA-256 and benchmark-only authority. CA-3 remains an unofficial copy; its existing source-specific rights limitations and broader Canada clearance state are unchanged.
• OD-002: exact bounded admission of ztl-v0.1 / 0.1.0, signed tag veraxis-ztl-input-v0.2-signed, commit 56e1ff0510c62b04dbd85bbe08b7a6deacbf276b, fixture-index SHA-256 ffadd65352d69ffcf55787c6dc26339e51eaed76b4c2ae789f7c813625247145. No ZTL runtime import, call, or semantic change.
• OD-003: Tier-1 reproduction is NOT ESTABLISHED and deferred to the experimental-release boundary; it is not represented as PASS.
• OD-004: non-executable ActionProposal → OIC exact proposal → RuntimeDecision → evidence handed downstream toward VEIP boundary. OIC has no VEIP lifecycle authority; VEIP does not reinterpret ZTL; absent integration fails closed; no adapter/import/execution is authorized.

Synthetic source identities

• SYN-NS-GOV-1: b882459931ddf06fb647e6885c1ced368b4a5f45b6441df618484a30473a2a24
• SYN-NS-PROC-1: cfb578274460cc17492f4b25a4db6893ce87a527e1207fe6a52b6384125ec073
• SYN-NS-AMEND-2: 8e197bc9e01554853163dfb5f84ad807893f678b26673737c8a72ddf0c8daf47
• retained CA-3: 6e89ad25847944ca2bd72bcbf02ec3d2942a234d373b6c10db44307e0fbdf2c3

Fail-closed evidence

The new stdlib-only verifier and contract tests refuse all 15 required conditions: synthetic-as-real authority, missing synthetic classification, byte mutation, provenance mismatch, fabricated government metadata, wrong ZTL commit/tag/index, stale currentness, Tier-1 escalation, executable VEIP integration, OIC lifecycle authority, VEIP reinterpretation of ZTL, premature semantic implementation, and global-manifest escalation.

Global verify-manifest --all remains truthfully INCOMPLETE / exit 3. Scoped prerequisite readiness does not launder global completeness.

Local producer verification

• git diff --check: clean
• make verify PYTHON=.venv/bin/python: schema 9/9, bootstrap 52/52, scoped verifier PASS, global manifest INCOMPLETE / exit 3 as expected
• ruff check .: PASS
• ruff format --check .: PASS (47 files)
• python -m mypy src tests: PASS (35 source files)
• python -m pip check: PASS
• python -m pip_audit --skip-editable: no known vulnerabilities
• SBOM generation: completed
• wheel build/smoke-install: completed
• forbidden-pattern scan: PASS
• Local pytest and make falsify: authoring host Python 3.12 process exited by signal 11; no assertion result was produced. GitHub-hosted exact-PR-composition CI is required and is the decisive suite evidence.

Claim ceiling

No semantic compiler correctness, source-interpretation correctness, legal correctness, benchmark performance, runtime authority, VEIP lifecycle, production-readiness, or expanded product claim is made.

SEMANTIC CODE-START GATE:
READY FOR SEPARATE EXACT-HEAD REVIEW
NOT OPEN

SEMANTIC IMPLEMENTATION HAS NOT STARTED

CI GREEN IS EVIDENCE, NOT ACCEPTANCE

NOT SELF-ADJUDICATED

Final GitHub-native evidence

• Final branch head: 4065e6a7c02badd3356f4c74be0815079d836aca
• Base: fc96d6f590a6f8ed12ecbb256f33f6de16d0c619
• GitHub synthetic merge composition: 309199520c84e5c9917de69c2d15e4a112717677
• Successful PR run: 33121629108
• All nine jobs succeeded: dependency-review, advisory-scan, bootstrap-integrity, schema-validation, lint, typecheck, test, sbom, compose-validation.
• Hosted suite: 1248 passed, 1 intentionally skipped (gitignored local receipts unavailable in CI), 1249 collected.
• Global manifest: INCOMPLETE / exit 3, one CA-3 row RECORDED_NOT_VERIFIED; not converted to PASS.
• Action runtime deprecation warnings: 0 observed.
• The compose job's “Show service logs on failure” step was skipped because all services became healthy; it is explicitly failure-only and inapplicable on success.

Preserved failure history and remediation

• 33104571953, head d98145f9e194a58b496852cf9e7b0333f9e965c5: test failed (12 failures). Root cause: scoped synthetic evidence had been added to frozen global SOURCE_MANIFEST.csv/STATUS.md, and a provenance filename collided with historical receipt-freeze assertions. Fix: restored global artifacts byte-for-byte, kept the new source set scoped, and renamed its record to PROVENANCE.json.
• 33121441202, head bb5c954074e2ee747ac0ec3d2179fb4428617285: test failed (2 failures). Root cause: historical tests categorically rejected all future docs/contracts/ changes despite this owner-authorized exact contract scope. Fix: replaced the blanket rule with an explicit four-file allowlist; all other contracts, draft schemas, adapters, and STATUS.md remain forbidden.
• 33121629108, head 4065e6a7c02badd3356f4c74be0815079d836aca: SUCCESS.

004A review-requested fail-closed correction

• Reviewed predecessor head: 4065e6a7c02badd3356f4c74be0815079d836aca
• Preserved review: OIC-SEMANTIC-CODE-START-REVIEW-001, review ID 5046228894, CHANGES_REQUESTED
• Defect: validate_evidence() refused a supplied semantic_paths list, but production load_and_validate() never discovered or supplied paths, so the real verifier could pass after an unauthorized src/oic addition.
• Correction commit/final head: 1b9755126f6df40f8a2799896c58c433a3c8ba89
• New synthetic merge composition: 42c427f4bee8df800fe9dbf379e71ee307fb8f70
• Final successor run: 33127281845 — SUCCESS
• 004A changed files only: scripts/verify_code_start_gate.py; tests/contract/test_semantic_code_start_gate_closure.py

The production verifier now uses git ls-files -- src/oic and compares the tracked tree to an immutable ten-path baseline derived from reviewed predecessor 4065e6a7c02badd3356f4c74be0815079d836aca. Added paths and missing/renamed baseline paths are passed by load_and_validate() to the existing fail-closed invariant. The baseline is a literal constant and is never regenerated from the current tree.

T1–T7 execute in the hosted suite and establish: exact baseline acceptance; refusal of semantic_parser.py; refusal of innocuously named helper.py; refusal of nested newmodule/engine.py; refusal of rename/substitution; production load_and_validate() detection without manual parameter injection; and refusal to self-admit a newly existing path.

All previously reviewed PASS dimensions remain untouched. OD-001 through OD-004, CA-3, synthetic bytes, ZTL tuple and semantics, Tier-1 disposition, VEIP boundary semantics, contracts/mapping, schemas, dependencies, workflows, claims, licensing, and runtime boundaries were not modified by 004A.

Hosted successor suite: 1255 passed, 1 intentionally skipped, 1256 collected. All nine jobs succeeded. Global manifest remains INCOMPLETE / exit 3. The requested-changes review remains visible and undismissed.

SEMANTIC CODE-START GATE:
READY FOR SUCCESSOR EXACT-HEAD RE-REVIEW
NOT OPEN

SEMANTIC IMPLEMENTATION HAS NOT STARTED

REVIEW-REQUESTED DEFECT CLOSED IN PRODUCER CANDIDATE

CI GREEN IS EVIDENCE, NOT ACCEPTANCE

NOT SELF-ADJUDICATED

[inventor1975]

OIC-SEMANTIC-CODE-START-REVIEW-001

Separate exact-head reviewer. Composition verified before substantive review.
No branch mutation: no edit, push, rebase, amend, merge, or auto-merge.

REVIEWED HEAD: 4065e6a
BASE: fc96d6f
SYNTHETIC MERGE: 3091995
STATE: DRAFT / OPEN / MERGEABLE / NOT MERGED · 19 changed files
FINAL CI: 33121629108, bound to this exact head, success

DISPOSITION: REQUEST CHANGES — bound to head 4065e6a

One defect, found by the §7 mandatory special attack. Everything else passed.

§7 special attack — verdict B: material fail-closed gap

validate_evidence(..., semantic_paths=...) correctly rejects a non-empty
list (scripts/verify_code_start_gate.py:138). But no caller ever supplies
that list:

• load_and_validate() (:141-165) calls validate_evidence() with seven
  arguments; semantic_paths stays None on every real run — the parameter
  is dead at the only production call site;
• the covering test (test_semantic_code_start_gate_closure.py:127-128)
  injects ["src/oic/semantic_parser.py"] BY HAND — it proves the refuser
  works, not that anything detects;
• searched independently for another mechanism: test_document_consistency's
  static scan targets zverify occurrences only; the historical scope
  control fences docs/contracts/, schemas/draft/, STATUS.md,
  adapters/ztl/ — not src/; semantic_conformance.py is a test-only
  fixture validator. src/oic/ already exists (10 non-semantic modules), so
  a new src/oic/semantic_parser.py crosses no fence.

Consequence, exactly as the work order framed it: after semantic
implementation lands under an undetected path, make verify still prints
PASS semantic code-start prerequisite evidence; gate remains NOT OPEN.
The gate's own guard cannot fire in production.

Minimal correction (no redesign): make load_and_validate() derive
semantic_paths mechanically — e.g. freeze the current ten src/oic/*.py
module names as the admitted non-semantic set and pass anything new or
renamed under src/oic/ (and any other owner-named prohibited roots) into
semantic_paths. The refuser and its test then stand unchanged.

Everything verified and clean

• §2 OD-001: all four source digests match byte-for-byte (SYN-NS-GOV-1
  b8824599…, SYN-NS-PROC-1 cfb57827…, SYN-NS-AMEND-2 8e197bc9…, CA-3
  6e89ad25… at its manifest path). Each synthetic file self-declares
  SYNTHETIC — FICTIONAL in its first line; issuer is a fictional
  "Northstar Test Institution"; the only dates are Created: stamps —
  no real-government effective date is invented. Stale vs current synthetic
  is explicit (SYNTHETIC_SUPERSEDED + superseded_by). CA-3 remains
  REAL_PUBLIC_CANADIAN_SOURCE_UNOFFICIAL_COPY, BOUNDED_TEST_INPUT_ONLY,
  effective state NOT_ESTABLISHED_NO_DATE_INVENTED, rights limitations
  unchanged. SOURCE_MANIFEST.csv and STATUS.md untouched (0 occurrences
  in the 19 changed files). Global status stays INCOMPLETE.
• §3 OD-002: profile ztl-v0.1, version 0.1.0, tag
  veraxis-ztl-input-v0.2-signed, commit 56e1ff05… — verified against the
  ZTL repository itself (the tag resolves to exactly that commit);
  fixture INDEX.json hashes to ffadd65352… exactly. Stale PR#18 pin is
  preserved as a correction (pin_correction_notice with superseded
  commit/tag), not rewritten. No ZTL runtime import/call anywhere in the
  diff; admission text is bounded to CODE START throughout.
• §4 OD-003: every Tier-1 mention in the diff reads NOT ESTABLISHED /
  deferred; the verifier itself rejects escalation
  ("Tier-1 state escalated" check). No PASS/COMPLETE/VERIFIED anywhere.
• §5 OD-004: boundary record carries all eight expected invariants
  verbatim, NON_EXECUTABLE_BOUNDARY_RECORD, no adapter, no lifecycle
  execution, no self-authorization.
• §6 Failure 33104571953: remediated by moving synthetic evidence to
  benchmarks/preflight/ — frozen artifacts restored (absent from the final
  diff), their tests strengthened rather than weakened. Failure 33121441202:
  remediated with a literal four-path allowed_contract_updates set
  subtracted from the blanket prohibition; everything else under
  docs/contracts/ (and STATUS.md, schemas/draft/, adapters/ztl/)
  stays fenced. This is a genuinely narrow owner-named exception, not a
  general escape hatch. One remark, not a blocker: the allowlist is
  permanent, so future mutation of those four files will pass the same
  fence; version-bump discipline on those v0.1-named files is what keeps
  that honest.
• §8: semantic implementation ABSENT. All 19 files inspected; src/oic
  on this head holds only baseline/cli/doctor/errors/hashing/manifests/
  paths/schemas — no parser, extraction, IR, OCE, Rego, OPA/Cedar/OpenFGA/
  Cerbos execution, ZTL/VEIP invocation, or benchmark result generation.
• §9 claim ceiling: respected; the gate document and OD-003 keep
  NOT OPEN / NOT ESTABLISHED; the verifier's success line itself restates
  "gate remains NOT OPEN".
• §10: all nine expected jobs SUCCESS on the exact head; the
  failure-only "Show service logs on failure" step is SKIPPED and correctly
  inapplicable — "Wait for every service to become healthy" succeeded.

Reviewer role and limitations

Produced by Vitaly Reznik's assistant; Vitaly Reznik is a collaborator of
the repository owner and the author of the admitted ZTL kernel. This is an
exact-head technical review — not independent third-party adjudication, not
legal review, not Tier-1 reproduction. Those categories remain separate and
unclaimed.

SEMANTIC CODE-START GATE — NOT OPEN BY THIS REVIEW.
SEMANTIC IMPLEMENTATION — NOT AUTHORIZED BY THIS REVIEW.
MERGE — NOT AUTHORIZED BY THIS REVIEW.

[inventor1975]

OIC-SEMANTIC-CODE-START-REVIEW-001 — SUCCESSOR EXACT-HEAD RE-REVIEW

Successor disposition: APPROVED — bound to exact head
1b97551 and CI run 33127281845 (9/9 SUCCESS
on that head; synthetic merge 42c427f).

The original CHANGES_REQUESTED review (id 5046228894, head 4065e6a…) is
preserved, not dismissed: it correctly described the predecessor.

Why the defect is closed — verified on the bytes, not the changelog

The successor differs from the rejected head by ONE commit touching exactly
TWO files (compared via the API: scripts/verify_code_start_gate.py +47/−1,
tests/contract/test_semantic_code_start_gate_closure.py +89/−0). Nothing else
moved, so every finding of the predecessor review carries over unchanged.

The dead parameter is dead no more:

• discover_unadmitted_production_paths() enumerates tracked src/oic
  paths via git ls-files and compares them against an immutable ten-path
  pre-gate baseline (frozenset in the verifier, stamped with the admitted
  pre-gate head). Additions AND missing paths are returned, so a rename
  surfaces as both.
• load_and_validate() now calls it and passes the result into
  validate_evidence(semantic_paths=...) on every production run — the
  §7 gap exactly.
• Absence of git is a GateEvidenceError, not a silent pass: fail-closed.

The attack tests are real, not hand-fed: they build git sandboxes and
exercise the actual mechanism — T1 exact baseline accepted; T2–T4 any new
tracked path (semantic_parser.py, innocuous helper.py, nested
newmodule/engine.py) refused; T5 rename/substitution refused; T6 proves
load_and_validate() detects WITHOUT any manual parameter — the precise
predecessor defect; T7 the working tree cannot self-extend the immutable
baseline.

Named limitations of the mechanism (not blockers)

• git ls-files sees tracked paths only. For exact-head review this is
  definitionally sufficient — an untracked file is not part of any head —
  but a locally-run make verify in a dirty tree will not see untracked
  semantic code.
• The scan covers src/oic only. A semantic implementation planted under a
  new root would not hit this detector; it would instead have to pass the
  bootstrap-integrity baseline manifest and the historical scope controls,
  which fence tree changes by a different mechanism. Widening the scan to
  owner-named prohibited roots remains available if the owner wants both
  fences on one mechanism.

Standing statements

Reviewer relationship unchanged and restated: produced by Vitaly Reznik's
assistant; Vitaly Reznik is a collaborator of the repository owner and the
author of the admitted ZTL kernel. Exact-head technical review only — not
independent adjudication, not legal review, not Tier-1 reproduction.

SEMANTIC CODE-START PREREQUISITES — APPROVED FOR OWNER GATE-OPENING DECISION.
SEMANTIC CODE-START GATE — NOT OPEN BY THIS REVIEW.
SEMANTIC IMPLEMENTATION — NOT AUTHORIZED BY THIS REVIEW.
MERGE — NOT AUTHORIZED BY THIS REVIEW.