Migrated from the pre-public development archive (issue #123 there); archive PR/issue numbers in the body refer to that archive.
Surfaced during the #113 fix (2026-07-21): re-verifying the live biome artifact after fixing OPT29 traded its FAIL for a DISTINCT pre-existing one (verify stops at the first failing finding, so this was masked).
The defect: check_saving_within_measured_compute's join binds a finding's job to the cost spine. OPT33 names its job by YAML key lint, but pull_request.yml's lint job appears in the spine under its name: override Lint project (...) (2259 + 10466 min/mo measured). The key lint misses that join, and the guard's cross-workflow same-name FALLBACK mis-binds it to an unrelated lint job in pull_request_markdown.yml (568 min/mo) → false FAIL ("credits 3318.8 … measures only 568"). OPT33's figure is actually within the real lint job's compute — the guard is wrong, not the finding.
Class: join fragility between findings' YAML-key job identity and the spine's display-name identity — affects the guard's by_job fallback and potentially any name:-overridden long pole. Related in spirit to #118 (check→job name resolution) but a distinct code path (the sizing-guard join, not the check mapper).
Fix direction: the join should resolve YAML key ↔ display name via the scanned job graph (both identities are known at collect time — stamp the mapping or join through it) and the cross-workflow same-name fallback should never win over a same-workflow key/name-resolved match. Needs its own fixtures (name-override shape) + the biome artifact as the live pin.
Live state: the biome scratch artifact still FAILs verify on this (expected, disclosed in PR for #113). Launch-relevant: same severity class as #112/#113 (gate false-positive on a real repo shape — name: overrides are extremely common).
Surfaced during the #113 fix (2026-07-21): re-verifying the live biome artifact after fixing OPT29 traded its FAIL for a DISTINCT pre-existing one (verify stops at the first failing finding, so this was masked).
The defect:
check_saving_within_measured_compute's join binds a finding's job to the cost spine. OPT33 names its job by YAML keylint, butpull_request.yml's lint job appears in the spine under itsname:overrideLint project (...)(2259 + 10466 min/mo measured). The keylintmisses that join, and the guard's cross-workflow same-name FALLBACK mis-binds it to an unrelatedlintjob inpull_request_markdown.yml(568 min/mo) → false FAIL ("credits 3318.8 … measures only 568"). OPT33's figure is actually within the real lint job's compute — the guard is wrong, not the finding.Class: join fragility between findings' YAML-key job identity and the spine's display-name identity — affects the guard's
by_jobfallback and potentially anyname:-overridden long pole. Related in spirit to #118 (check→job name resolution) but a distinct code path (the sizing-guard join, not the check mapper).Fix direction: the join should resolve YAML key ↔ display name via the scanned job graph (both identities are known at collect time — stamp the mapping or join through it) and the cross-workflow same-name fallback should never win over a same-workflow key/name-resolved match. Needs its own fixtures (name-override shape) + the biome artifact as the live pin.
Live state: the biome scratch artifact still FAILs verify on this (expected, disclosed in PR for #113). Launch-relevant: same severity class as #112/#113 (gate false-positive on a real repo shape —
name:overrides are extremely common).