[drk-ram]

What this changes

Closes #625. Current source head: b9cc53379ba8e5eee222e42f48189734497fbbf3.

conformance/src/trace_tests/inclusion.py rejects unpaired UTF-16 surrogates in values and keys, non-string dictionary keys, and circular containers while preserving shared acyclic containers. parse_proof requires exact integer indices and counts within the safe-integer range, a positive count, and an index below the count when supplied.

parse_entry rejects blank identifiers and ASCII control characters, including DEL, in producer and batch_id. It accepts internal spaces, including producer="Example Gateway 1.0"; the earlier description incorrectly said all whitespace was rejected. leaf_count must be an exact integer between 1 and 2**53 - 1.

The draft anchor-profile text documents the surrogate exclusions, integer bounds, and identifier rules. spec/trace-v0.2.md is unchanged. Existing anchor vectors keep their verdicts; malformed index/count inputs are rejected at the parsing boundary.

The latest review recommended narrowing identifier rejection to blank strings and ASCII controls. The current source and draft text implement that recommendation; this description now matches them and corrects the review-bar count from 16 to the measured 20.

Type of change

• Editorial (typo, link fix, clarification: no normative effect)
• Non-breaking spec change (new optional field, new platform profile, informative addition)
• Breaking spec change (requires 14-day comment period and Project Lead sign-off)
• Schema change
• Example addition

Reference conformance implementation and draft companion-profile clarification. Final governance classification remains with the maintainers; the published TRACE v0.2 specification is not modified.

Spec section

spec/registry-anchor-v1.md sections 1, 4, and 5. No edit to spec/trace-v0.2.md.

Tests for a fix

tests/test_conformance_inclusion_profile_and_receipt_boundaries.py covers surrogate values/keys, non-string keys, cycles versus shared acyclic containers, index/count types and bounds, blank/control-containing identifiers, unsafe counts, and acceptance of an internal space. Its separate test_parse_proof_without_leaf_count_rejects_leaf_index_above_safe_int pins the upper index bound for proof-only receipts.

Rerun on the current head with Python 3.11.9, importing this checkout's src and conformance/src through PYTHONPATH:

• python -m pytest tests\test_conformance_inclusion_profile_and_receipt_boundaries.py -q --tb=short -> 20 passed.
• From conformance: python -m pytest tests\unit\test_tr_anc.py tests\test_tr_anc_inclusion.py tests\test_anchor_vectors.py -q --tb=short -> 81 passed.
• python tools\review_bar.py --base origin/main --head HEAD -> 20 fail on behaviour, 0 fail on a missing name or file, 0 pass or skip, measured against merge base b938790. This is contribution-level failing-before evidence, not a claim of independently isolated mutation coverage for every clause.

The earlier combined root/conformance pytest command hits a tests.conftest package-name collision in this Windows environment; running the two packages separately succeeds. Hosted CI and Conformance CI on this head are successful. Maintainer approval is still required.

Changelog entry

• Conformance (trace_tests.inclusion): reject unpaired UTF-16 surrogates, non-string dictionary keys, and circular references; validate safe-integer leaf_index and positive leaf_count bounds; reject blank identifiers and ASCII control characters while preserving internal spaces (conformance(inclusion): enforce registry-anchor-v1 surrogate/key/cycle rules and validate leaf_index/entry fields #625, by @drk-ram).

Checklist

• DCO sign-off on all commits (git commit -s)
• Breaking changes marked with <!-- CHANGED: #NNN: description --> in spec text
• Backward compatibility statement included (for breaking changes)

[github-actions]

🟡 Contributor Check: MEDIUM

Check	Result
Profile	MEDIUM
Credential	LOW
Overall	MEDIUM

Automated check by AgenTrust Contributor Check.

[rajnisht7]

the overall changes looks good, the cycle detection, iterative traversal, and boundary checks look solid, and the tests cover the important cases well.

One thing: a few of the new rules are currently stricter than the published normative text in registry-anchor-v1.md.

• §1 does not yet explicitly prohibit surrogate code points (U+D800–U+DFFF) in strings/keys
• §4 only requires leaf_count ≥ 1; it does not state the ≤ 2**53-1 bound (same for leaf_index).
• §4 defines producer / batch_id simply as strings; it does not currently forbid whitespace or control characters

it will be good to update the normative text in registry-anchor-v1.md so the published requirements match what the conformance suite now enforces, once the spec and implementation agree, this looks good to land

non-blocking

• also positive 2**53-1 boundary tests and a plain pytest.raises for the cycle can be added

[lywinged]

Request changes. Reviewed at 19c65b2 on da69b31.

This change makes a conformance test fail, so, as CONTRIBUTING's "Before a pull request is reviewed" asks, this first review stops at what is missing. The new character check in parse_entry loops over _ENTRY_STRINGS, which includes ts: conformance/tests/test_tr_anc_subject.py sets ts to "2026-10-16 00:00:00Z", expects "ts must be an ISO-8601 UTC timestamp", and now gets "registry entry ts must be a non-empty string without whitespace or control characters". It fails here on 3.11 and 3.12, and Conformance CI failed in its full-suite step on 3.12. The validation command in the description does not run that file.

The same section asks for a test for each rejection the diff adds, failing when that rejection is switched off, or a sentence saying why no test can observe it. Three have neither: the leaf_index upper bound, the 1 <= on the receipt's leaf_count and the 0x7F check. I will review the rest once these are in.

CI at 19c65b2 also failed in the portability step, for a reason this pull request did not cause: main failed the same step at da69b31, before #624, and passes at dcca4e0.

---

Generated by Claude Code

[drk-ram]

Thank you @rajnisht7 and @lywinged! Addressed all points in 8f9ca2b (rebased onto origin/main at e9ab302):

1. Scoped the whitespace/control-character check in parse_entry to producer and batch_id:
   • Left ts validation after the non-empty str check to _is_utc_timestamp(entry["ts"]) so conformance/tests/test_tr_anc_subject.py (ts = "2026-10-16 00:00:00Z") raises "registry entry ts must be an ISO-8601 UTC timestamp ending in Z, ..." and passes all tests.
2. Isolated tests for leaf_index upper bound, 1 <= on leaf_count, and 0x7F:
   • Added test_parse_proof_without_leaf_count_rejects_leaf_index_above_safe_int (parse_proof({"leaf_index": 2**53, "audit_path": []})), which fails if <= _SAFE_INT on leaf_index is removed.
   • Pinned match="receipt leaf_count must be an integer >= 1" on leaf_count: 0 and leaf_count: -1 in test_parse_proof_and_receipt_reject_invalid_leaf_index_or_count, which fails if 1 <= on leaf_count is removed.
   • Added 0x7F ("prod\x7f1.0" and "batch\x7f1") and positive 2**53 - 1 boundary assertions to test_parse_entry_rejects_whitespace_control_chars_and_unsafe_leaf_count.
3. Aligned spec/registry-anchor-v1.md §1, §4, and §5:
   • Updated §1 (unpaired UTF-16 surrogates), §4 (1 <= leaf_count <= 9007199254740991, non-empty producer / batch_id without whitespace or ASCII control characters U+0000 to U+001F, U+007F), and §5 (0 <= leaf_index <= 9007199254740991).

[lywinged]

Request changes. Reviewed at 8f9ca2b on e9ab302.

What the first round asked for is in. The whitespace and control-character check in parse_entry runs over producer and batch_id only, and the conformance suite passes under the default locale and under LC_ALL=C PYTHONUTF8=0 on 3.11 and 3.12, 974 passed, 2 skipped and 5 xpassed. The leaf_index upper bound, the 1 <= on the receipt's leaf_count and the 0x7F check each have tests that fail when the clause is removed.

The one thing is the normative text. The push rewrote registry-anchor-v1.md in three places, and each makes something malformed that the published text allowed and main accepts. Section 1 excludes an unpaired UTF-16 surrogate in a string or key: canonical_claim_bytes on main writes it as \ud800, this head refuses it. Sections 4 and 5 bound leaf_count and leaf_index at 9007199254740991: parse_entry on main accepts a leaf_count of 2**53, this head calls it malformed. Section 4's meaning column now says producer and batch_id carry "no whitespace or ASCII control characters": "producer": "Example Gateway 1.0" passes TR-ANC-002 on main and is a malformed registry entry here. I would keep the first two, which spell out what a UTF-8 implementation cannot represent and the range section 1 already imposes on claims, and I would narrow the third, in the code and in the text alike, to blank strings and ASCII control characters, since the published text allowed a space and a control character has no place in an identifier. Whether the whitespace rule is a breaking change under GOVERNANCE, which asks for a 30-day comment period, a backward-compatibility statement and Project Lead sign-off where the template's line says 14 days, I read as no: that process is written for spec/trace-v0.2.md and exempts drafts, and the changed cells carry no RFC 2119 keyword. The diff carries no <!-- CHANGED --> marker either way. Whichever rule section 4 ends with, the suite wants a test for it: swapping ch.isspace() for a blank check leaves both suites green, so the inner-space refusal, the one rule the control-character clause does not already cover, has none.

What I checked rather than read:

• CI's commands, run here on 3.11 and 3.12: ruff and mypy as CI runs them, which do not reach conformance/, and the dash check clean; the suite at 5240 passed and 98 skipped on both, with the head's conformance package on the path as CI installs it; the replay experiment, the two LC_ALL=C PYTHONUTF8=0 runs, the PYTHONOPTIMIZE=2 command and the two --help calls pass. 3.13, which Conformance CI also runs, is not installed here.
• The 27 anchor-inclusion vectors keep their verdicts here and on main through the TR-ANC check. Vectors 17 to 19, the out-of-range, negative and boolean index, now fail as a malformed receipt rather than a root mismatch; vector 20, the empty tree, fails as a malformed receipt rather than a malformed entry.
• Each clause named in the first round, removed alone, in the suite CI runs as pytest: the leaf_index upper bound fails 2 tests, the 1 <= on leaf_count 2, the 0x7F check 2. The whole producer and batch_id check fails 5 there and 3 more in the conformance suite, and type(leaf_index) is int loosened to isinstance fails 1.
• On main, parse_entry accepts a blank producer, an inner space, a non-breaking space or DEL in producer, and a tab or NUL in batch_id; this head refuses all six, and ts with a space gets the timestamp message on both.
• tools/review_bar.py reports all 20 cases of the new file under tests/ failing on behaviour at the merge base; the 16 cases the diff adds under conformance/tests are outside what the tool reads. The description's validation line still says 16 for the tool.

Things I would not hold the merge for:

• Section 1's exclusions paragraph is also rewritten by #537, differently, so whichever of the two merges second will have to resolve that paragraph.
• A proof-only receipt, the section 5 shape, with leaf_index 1 against a one-leaf entry still reads as a root mismatch; the same receipt with leaf_count 1 is now refused as malformed, so the two shapes answer differently.

---

Generated by Claude Code

[drk-ram]

@lywinged Rebased onto origin/main (b938790) in b9cc533, narrowed the producer and batch_id rule in parse_entry and spec/registry-anchor-v1.md §4 to reject blank strings (not val.strip()) and ASCII control characters (U+0000–U+001F, U+007F) while preserving internal spaces such as "Example Gateway 1.0", added that positive control to the test suites, and updated the PR description's tools/review_bar.py count to 20.

[drk-ram]

The current head b9cc533 follows the recommended identifier rule: blank strings and ASCII controls are rejected, but internal spaces are accepted. I corrected the description and changelog proposal, and replaced the stale review-bar count with the measured 20 behavioral failures at the merge base. Validation: 20 boundary tests and 81 focused conformance tests pass. Please re-review the remaining changes request.