[drk-ram]

What this changes

Contributor: Raghuvar Durga (@drk-ram)

Closes #557

Fixes two defects in conformance/src/trace_tests/report.py and preserves record_digest as a total diagnostic identifier across all records per maintainer review:

1. Include "TR-APR" in report.LEVEL_MODULES[0]: Updates LEVEL_MODULES[0] from ("TR-ENV", "TR-SIG", "TR-POL") to ("TR-ENV", "TR-SIG", "TR-POL", "TR-APR") so the report module breakdown matches runner._LEVEL_MODULES[0] and cumulative level schedules across Levels 0, 1, and 2.
2. Guard non-dict record["trace"] in _assemble: Checks isinstance(raw_trace, dict) before reading trace.get("transparency") so malformed CMCP records where record["trace"] is a non-dict ("not-a-dict", 123, None, [1, 2], True) report transparency=None (unanchored in HTML) instead of crashing report.build with AttributeError.
3. Retain total diagnostic record_digest across all report surfaces: Per @imran-siddique's ruling following @rajnisht7's review (and companion commit 783f06d by @dinakarjs), record_digest continues to hash sorted-key ASCII JSON for every record (including failing records with floats, NaN, Infinity, or out-of-range integers) without a separate fallback regime, ensuring record_digest(record), report.build(...).digest, to_json, and to_html always produce the same digest.

Self-review notes:

• Updated per @rajnisht7 and @imran-siddique's review to drop canonical_claim_bytes from record_digest and remove the second fallback hash in _assemble, so record_digest remains a total diagnostic identifier for all records (including failing boundary records) and agrees across record_digest(), build(), to_json(), and to_html().
• Replaced em dashes in conformance/src/trace_tests/report.py with -- so python tools/check_dashes.py passes after modifying report.py.

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

Spec section

None (conformance/src/trace_tests/report.py, conformance/tests/test_report.py, and tests/test_conformance_report_parity.py).

Tests for a fix

Added tests/test_conformance_report_parity.py and corresponding tests in conformance/tests/test_report.py. Running python tools/review_bar.py --base origin/main --head HEAD against the merge base (63e38c6) reports 6 failing on behaviour and 0 on missing names (plus 6 verifying that record_digest, build, to_json, and to_html agree on out-of-profile diagnostic records):

• tests.test_conformance_report_parity::test_report_level_modules_include_tr_apr_at_level_zero (fails on main with AssertionError: assert ('TR-ENV', 'TR-SIG', 'TR-POL') == ('TR-ENV', 'TR-SIG', 'TR-POL', 'TR-APR'))
• tests.test_conformance_report_parity::test_report_build_guards_non_dict_trace[str|int|none|list|bool] (5 cases; all fail on main because report.build raises AttributeError on non-dict record["trace"])
• tests.test_conformance_report_parity::test_report_record_digest_remains_diagnostic[nan|inf|neg-inf|float|above-safe-int|below-safe-int] (6 cases; pins single-regime agreement across record_digest, report.build, to_json, and to_html for out-of-profile values)

Validation commands and results:

• pytest tests/test_conformance_report_parity.py -v -> passed (12 passed)
• pytest conformance/tests/test_report.py -v -> passed (31 passed)
• pytest conformance/tests -q -> passed (791 passed, 5 xpassed)
• python tools/review_bar.py --base origin/main --head HEAD -> passed (6 fail on behaviour, 0 fail on a missing name or file, 6 pass)
• python tools/check_dashes.py -> passed (no dashes outside examples/, spec/trace-v0.1.md)
• ruff check conformance/src/trace_tests/report.py conformance/tests/test_report.py tests/test_conformance_report_parity.py -> passed (All checks passed!)
• mypy --strict conformance/src/trace_tests/report.py tests/test_conformance_report_parity.py -> passed (Success: no issues found in 2 source files)

Checklist

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

[rajnisht7]

LEVEL_MODULES parity and the non-dict trace guard looks correct.
however the issue is record_digest, it now goes through canonical_claim_bytes, so floats / NaN / unsafe integers raise InclusionError. _assemble catches that and falls back to a different json.dumps(..., allow_nan=True) digest so build() still works, good for reportability, but it leaves two digest regimes for the same record:

record_digest(record) -> raises
report.build(...).digest -> fallback hash

this report still tells readers to “compare the digest you compute,” but for boundary records they cant get that value from record_digest() and the docstring also don’t document the new strict contract or the fallback

so either:

• keep record_digest as a total diagnostic identifier and use canonical_claim_bytes only where anchor admissibility is required, or
• Document strict record_digest (InclusionError out of profile), treat the _assemble fallback as diagnostic-only, and add e2e tests that build / HTML / JSON succeed with a deterministic fallback digest for NaN, 1.5, and 2**60, and that this isn’t presented as a registry-anchor digest.

also worth noting:

• The fallback uses allow_nan=True; if you keep it, state clearly that it’s diagnostic-only and not registry-anchor canonicalization.
• A separate digest_profile (or similar) would make the two regimes explicit for readers.
• The non-dict trace test only catches AttributeError; asserting the report fields is enough for now, but a broader “must not raise” would be slightly stronger.
• Existing obligation-accounting boundary tests stay green via the fallback; the gap is the public record_digest contract and e2e coverage of that fallback path.

[imran-siddique]

Thanks @drk-ram, and thanks @rajnisht7 for laying out the two options. Let's take option 1 and keep record_digest working for every record, as it does today.

The report is there to describe records that fail, including ones with floats, NaN or very large numbers. If record_digest throws on those, the report needs a second fallback hash, and a reader ends up with two different digests for the same record.

So please drop the switch to canonical_claim_bytes in record_digest, and keep the other two fixes: TR-APR in LEVEL_MODULES[0] and the guard for a trace that is not an object. Those look good.

[dinakarjs]

Exact-head check of c10ff01: Imran's diagnostic-digest adjustment is not yet present. Companion for @drk-ram to cherry-pick into this original PR, retaining ownership: dinakarjs@783f06d . It preserves TR-APR parity and the non-dict trace guard, restores the original diagnostic serialization, removes the second fallback regime, and tests public digest/build/HTML/JSON agreement for NaN, infinities, fractions and unsafe integers.

Validation on the companion changes, with exact checkout imports: full conformance 791 passed, 5 xpassed after building/installing this checkout's wheel into an isolated metadata target; 12 root parity tests passed; ruff passed. Independent cold read found no blocking issue; its clearer-test-name suggestion is incorporated. Creating the companion PR directly into the author's fork returned GitHub 403, so handing off the DCO-signed commit here. This is not approval of the unchanged current head; no merge requested.

[drk-ram]

Thank you @imran-siddique, @rajnisht7, and @dinakarjs!

I have cherry-picked @dinakarjs's companion commit (6465366, from 783f06d82b52670338fe63a2edc72885186af198) implementing Option 1 per @imran-siddique's ruling:

• Dropped the switch to canonical_claim_bytes in record_digest and removed the fallback hash in _assemble, keeping record_digest as a single total diagnostic digest across record_digest(), report.build(), to_json(), and to_html() for all records (including failing boundary records with floats, NaN, Infinity, or unsafe integers).
• Retained "TR-APR" in report.LEVEL_MODULES[0] and the non-dict record["trace"] guard in _assemble.
• Added end-to-end tests verifying that record_digest, report.build, to_json, and to_html agree on out-of-profile values (nan, inf, -inf, 1.5, 10**20, -10**20).

[dinakarjs]

Adoption verified at exact head 6465366. Imran's selected single diagnostic-digest contract is implemented. TR-APR parity and the non-object trace guard remain, while digest/build/JSON/HTML agree for NaN, infinities, fractions and unsafe integers without a second fallback regime.

With this head's hash-locked root/conformance dependencies and explicit checkout import paths, all 791 conformance tests pass with five xpasses; the 12 root parity controls also pass. These suites run separately because their tests packages collide when collected together. Restoring strict canonical-claim hashing only in memory makes six diagnostic-boundary controls fail while the other six controls pass. Ruff, targeted strict mypy for report.py and the root parity module, root-library mypy, dash and whitespace checks pass. CI and Conformance CI succeed; ClusterFuzzLite was still running when checked.

No additional blocking finding in this adoption check. I authored the companion digest correction, so this is incorporation/regression evidence rather than an independent maintainer approval of my own changes. Diagnostic identity remains distinct from registry-anchor admissibility; five xpasses are not five additional acceptance requirements.

[rajnisht7]

changes looks good, LEVEL_MODULES[0] now matches the runner, non-dict trace is guarded cleanly, and record_digest stays a single diagnostic path as ruled. Tests cover the cases.

[imran-siddique]

Thanks @drk-ram and @dinakarjs. This is option 1 from 17 October: record_digest stays one digest for every record, with no fallback hash, and the TR-APR and non-object trace fixes are kept. @rajnisht7 your approval was dismissed only by the rebase; the two test files are unchanged and report.py differs from your approved head only by what main brought in.