[drk-ram]

What and why

Closes #146.

Fixes four verification and registry-anchor-v1 parity bugs across src/trace_verify/ and tools/:

1. Enforce registry-anchor-v1 section 1 profile rules in canonical_claim_bytes and anchor_profile_violation:
   • Check for unpaired UTF-16 surrogates (U+D800..U+DFFF) in string values and object keys, detect circular references (active: set[int]), and walk (list, tuple) in anchor_profile_violation.
   • Call anchor_profile_violation inside canonical_claim_bytes across src/trace_verify/_verify.py, tools/verify_inclusion.py, tools/anchor.py, and tools/batch_anchor.py, raising ValueError on floats, integers outside [-9007199254740991, 9007199254740991], unpaired surrogates, circular references, and RecursionError / TypeError, while preserving the explicit as-transmitted contract (raw_bytes returned verbatim without re-serialization).
2. Reject private/symmetric JWK parameters and non-canonical base64url x:
   • Reject RFC 7517 / RFC 7518 / RFC 8037 private or symmetric JWK members (d, p, q, dp, dq, qi, oth, k) and non-canonical 32-byte base64url x encodings (including non-zero trailing padding bits) in src/trace_verify/_signature.py (validate_public_key_jwk / verify_claim_signature) and tools/validate_producers.py.
3. Require exact int in MMR proof deserialization, builders, and verifiers:
   • Enforce type(v) is int (rejecting bool, IntEnum, str, and float) in InclusionProof.from_dict, ConsistencyProof.from_dict, inclusion_proof, consistency_proof, verify_inclusion, and verify_consistency in src/trace_verify/_mmr.py.
4. Reject duplicate JSON member names across standalone, validation, and NDJSON tools:
   • Use loads_unique (object_pairs_hook) in tools/verify_inclusion.py, tools/verify_checkpoint_chain.py, tools/validate_registry.py, tools/validate_producers.py, tools/anchor.py (_registry_is_chained), tools/batch_anchor.py (is_already_anchored, load_published_entries), and src/trace_verify/_signature.py.

Validation

• python -m unittest discover -s tests -q — passed (416 tests)
• python -m pytest tests/test_mmr_core.py tests/test_cli_duplicate_members.py tests/test_trace_verify_pkg.py tests/test_signature.py -q — passed (95 passed, 74 subtests passed)
• python -m ruff check src/ tools/ tests/ — passed (All checks passed!)
• python tools/validate_registry.py — passed (OK: 3 day file(s) valid)
• python tools/validate_producers.py — passed (OK: 3 producer key file(s) valid)

Registry impact

• This PR does not modify any existing line under registry/

Existing entries are append-only. If you are changing one, say why here: mirrors
and auditors may already have observed the current commit hashes, and a rewrite
diverges them.

Checks

• pytest passes
• Registry and producer files still validate (tools/validate_registry.py,
  tools/validate_producers.py)
• If the anchor format or verification semantics changed, docs/anchor-format.md
  is updated in the same PR, and existing inclusion proofs still verify
• Commits are signed off (DCO)

[dinakarjs]

Reviewed exact head f1ea2ff (full commit attached to this review). Checked strict MMR integer/proof boundaries, the public-JWK/base64url restrictions, duplicate-member parsing in the CLI/tool paths, and sorted-key anchor profile validation without replacing the separate signing canonicalizer. Local full unittest suite: 415 tests passed; Ruff passed. Hosted CI and CodeQL succeed. ClusterFuzzLite remains action_required and is not represented as executed evidence. No blocking finding.

[rajnisht7]

the fixes looks good, sorted-key profile checks, private JWK + canonical x, MMR exact-int, and loads_unique on the four named tools all check out. Scripts and new tests pass on this head.

One thing needs to be addressed:

• Profile walk now runs before json.dumps, but there’s no cycle detection. A self-referential dict hangs instead of failing closed. kindly detect cycles by identity, return a clear error, allow shared-but-acyclic objects, and add tests for cyclic dict/list in both _verify.py and tools/verify_inclusion.py.

Note: as-transmitted still returning raw_bytes without a profile check is consistent with the docs and existing design (exact bytes, no re-serialization). Tool intake already rejects out-of-profile claims, and we already have coverage that as-transmitted preserves exact bytes including whitespace/key order.

Follow-up (non-blocking):

• Surrogate wording can stay aligned with “unpaired” (behavior is fine in Python).
• is_already_anchored / load_published_entries / _registry_is_chained still use plain json.loads on published NDJSON; loads_unique there would match the hardening but is outside the #146 scope.

[drk-ram]

Thanks @rajnisht7! Addressed in the latest commit: anchor_profile_violation in both src/trace_verify/_verify.py and tools/verify_inclusion.py now tracks the active ancestor object IDs on the explicit traversal stack (active: set[int] with (value, where, exiting) frames) so cyclic dicts/lists fail cleanly with a "circular reference at ..." violation while shared-but-acyclic dicts/lists continue to succeed, covered by unit tests in tests/test_trace_verify_pkg.py.

[rajnisht7]

Cycle detection looks correct, active-set + exiting frames in both copies, shared-but-acyclic objects still allowed, clear error. Tests exercise both implementations.

needs follow up (non blocking):

• as-transmitted path still returns raw_bytes without the profile walk (consistent with the existing design, but the contract should be made explicit in comments/tests).
• is_already_anchored / load_published_entries / _registry_is_chained still use plain json.loads on published NDJSON; switching them to loads_unique would make the trust boundary consistent.
• Add an explicit regression test that invokes tools/validate_producers.py for duplicate members.
• inclusion_proof() should enforce type(size) is int at the top (IntEnum currently passes).

[drk-ram]

Thanks @rajnisht7! Addressed all four follow-up items in fbb2c1e: (1) enforced type(size) is int at the top of inclusion_proof() in src/trace_verify/_mmr.py (with IntEnum regression coverage in tests/test_mmr_core.py); (2) switched is_already_anchored, load_published_entries, and _registry_is_chained in tools/batch_anchor.py and tools/anchor.py to use loads_unique; (3) made the as-transmitted exact-raw_bytes contract explicit in comments across src/trace_verify/_verify.py, tools/verify_inclusion.py, tools/anchor.py, and tools/batch_anchor.py and in test_as_transmitted_returns_raw_bytes_without_reserialization (tests/test_trace_verify_pkg.py); and (4) added an explicit regression test in tests/test_cli_duplicate_members.py invoking tools/validate_producers.py on a duplicate-member producer key file alongside the NDJSON helpers. Full unittest discovery (416 tests) and Ruff pass.