CNA-BUG-247: The EasyGL L7 policy and report justify zero tolerance with 'all 137 renderable assets' while the report's own fields record 140 captured, byte-identical assets

CNA snapshot 3b43bd86  ·  Known Issues › Current bugs  ·  source links pinned to 3b43bd86

The EasyGL L7 policy's justification, copied into its report, cites 137 renderable assets while the report records 148 assets, 140 byte-identical captures and 8 rejections.

Identifier

CNA-BUG-247

Category

Bug

Subsystem

Documentation & release tooling

Status

Open

Verified against

CNA 3b43bd86 (3b43bd8687ed46523d873f9bec4b32bd8481dcb5)

Severity

Low (a triage suggestion, not a project priority)

Evidence basis

Source-verified: read at TARGET, not executed

Tests touching this area

Yes: see Current tests

Affected contract

tests/gltf-l7/easygl-policy.json and docs/gltf-l7-corpus-report.json goldenComparison.justification (the OPENGLES3/EasyGL L7 tolerance rationale)

Expected behaviour

The hand-written rationale agrees with the machine fields it justifies, as docs/gltf-conformance.md does (“148 explicit EasyGL dispositions: 140 PNGs + 8 safe rejections”). The SOFTWARE policy phrases the same rationale without a count (“every renderable asset”).

Actual behaviour at TARGET

easygl-policy.json says two independent llvmpipe/EasyGL processes produced byte-identical PNGs “for all 137 renderable assets”; gltf-l7-corpus.py copies the policy's goldenComparison into the report, so gltf-l7-corpus-report.json carries the same sentence beside distinctAssetCount 148, capturedAssetCount 140 and rejectedAssetCount 8; its assets array has 140 capture dispositions, each with twoProcessPngByteIdentical: true, and 8 reject. A planning document that repeated “137 deterministic PNGs plus 8” was deleted on 2026-10-11 (96c6e77be). The figure 137 matches earlier campaign rows in plans/plan_gltf.md (GLTF-390, GLTF-471), which suggests the sentence was carried over rather than regenerated.

Source locations

• tests/gltf-l7/easygl-policy.json — goldenComparison.justification
• docs/gltf-l7-corpus-report.json — goldenComparison.justification against distinctAssetCount, capturedAssetCount, rejectedAssetCount and assets
• scripts/gltf-l7-corpus.py — the report's goldenComparison is the policy's, copied
• docs/gltf-conformance.md — L7 row: 140 PNGs plus 8 safe rejections
• tests/gltf-l7/software-policy.json — count-free phrasing of the same rationale

Evidence

Checked by reading at 3b43bd86 (the JSON files were parsed to count dispositions; no capture was run). The justification sentence is identical in the policy and the report; gltf-l7-corpus.py builds the report's goldenComparison from the policy. Whether the 137 figure was ever true of this report cannot be decided from the snapshot; only its disagreement with the report's own fields is established. Hand-written prose lagging generated fields, not a rendering defect.

Focused reproduction

No focused reproduction is known. Nothing has been invented here; the evidence above is what exists.

Current tests

GltfFixtureCorpusTests.cpp validates the SOFTWARE L7 report for completeness and reproducibility (SoftwareCorpusL7ReportIsCompleteExactAndReproducible); no test reads the EasyGL report, and none compares a justification's prose with the counts.

Regression test

Drop the count from the EasyGL justification (as the SOFTWARE policy does) or have gltf-l7-corpus.py refuse a justification whose stated count differs from capturedAssetCount.

Blast radius

Reviewers auditing the EasyGL L7 evidence and readers of the project status note; the gate's zero tolerance and its 140 goldens are unaffected.

Workaround

Use the report's machine fields (and the site's L7 summary) rather than the justification's count.