[synth coordinator — found by the v0.69.0 clean-room review]
scripts/artifact_citation_check.py:117 globs artifacts/*.yaml — non-recursively. Measured on the v0.69.0 tree (de049f24):
artifacts/*.yaml 30 files <- what the gate scans
artifacts/**/*.yaml 127 files
v0.69 artifacts seen 0
Every release from v0.61 on keeps its artifacts in a subdirectory, so all 97 subdirectory files — including all nine v0.69 artifacts — are invisible to it.
The gate also reads only keys literally named run:, and no release artifact at any version has ever had one, so all 12 cited filters it actually checks come from e2e-verification.yaml, sw-verification.yaml and verification.yaml.
The v0.69 artifacts do cite tests — --test every_section_header_is_named_1307, --test mpu_honest_1284, --lib ra003 — and none was checked. The gate's green says nothing about the release it ran on.
The precedent is already in this repo: status_evidence_check.py carries an explicit P3 rule for exactly this class ("yaml under artifacts/ that PROGRAMME_GLOB does not scan — invisible to every P-rule"). The citation gate has no equivalent.
The fix must be red-first. A citation to a test that does not exist, placed in a release-subdirectory artifact, must red the gate before the glob is widened — otherwise widening it is unfalsifiable.
Scoped as RQ-70-CITEGAP.
[synth coordinator — found by the v0.69.0 clean-room review]
scripts/artifact_citation_check.py:117globsartifacts/*.yaml— non-recursively. Measured on the v0.69.0 tree (de049f24):Every release from v0.61 on keeps its artifacts in a subdirectory, so all 97 subdirectory files — including all nine v0.69 artifacts — are invisible to it.
The gate also reads only keys literally named
run:, and no release artifact at any version has ever had one, so all 12 cited filters it actually checks come frome2e-verification.yaml,sw-verification.yamlandverification.yaml.The v0.69 artifacts do cite tests —
--test every_section_header_is_named_1307,--test mpu_honest_1284,--lib ra003— and none was checked. The gate's green says nothing about the release it ran on.The precedent is already in this repo:
status_evidence_check.pycarries an explicit P3 rule for exactly this class ("yaml under artifacts/ that PROGRAMME_GLOB does not scan — invisible to every P-rule"). The citation gate has no equivalent.The fix must be red-first. A citation to a test that does not exist, placed in a release-subdirectory artifact, must red the gate before the glob is widened — otherwise widening it is unfalsifiable.
Scoped as
RQ-70-CITEGAP.