Repository navigation
Collect audit.sh accuracy reports: false positives / false negatives #19
Description
What
A standing issue to collect audit.sh misfires from real repos:
- False positive: audit says a file/section is missing when it exists in a nonstandard location
- False negative: audit passes something that is actually a placeholder or empty stub
How to report
Comment with: repo shape (public link or description), the audit line, expected vs actual. Each confirmed misfire becomes its own fix issue.
Why
audit.sh scoring is the product's first impression; accuracy reports from strangers' repo shapes are the highest-value contribution a new contributor can make. Good first issue: reproducing a report against the fixtures in tests/ is a self-contained task.
Activity
AnayDhawan commentedon Aug 14, 2026
Self-audit pass, 2026-08-14. Staying open by design, per the issue's own framing as a standing collection point. Logging the run here so the next one has a baseline to diff against.
Re-ran
bash scripts/audit.sh <path>against the same three repos used as the pre-v1.0.0 dogfooding proof (recorded in the maintainer's decision log for 2026-07-08, at Components 93% / tourneyradar 93% / studymap 81%):AnayDhawan/ComponentsAnayDhawan/tourneyradarStudentSuite/StudyMapScoring is stable across the seven weeks: no drift, no regression, and the one delta is explained by the audited repo changing rather than by
audit.shchanging.Two genuine misfires found and filed as their own issues, per the reporting mechanism above:
new_component.yml(correct for a component registry).check_anyhardcodesfeature_request.yml/.md, so it reads as absent and costs a scored point. Components is really 16/16.<!-- DEMO GIF: … belongs here -->TODO comment that satisfies the demo check by itself. Inverse case: StudyMap's livecontrib.rockscontributors badge fails the badge check. Also notes that the badge check is scored but rendered as a yellow⚠, so it looks report-only while costing a point.Both are reproducible against fixtures in
tests/and labelled good first issue.Nothing else misfired: the remaining
✗marks on StudyMap (no dependabot config, no demo image) were verified against the filesystem and are true negatives.AnayDhawan commentedon Aug 16, 2026
Follow-up on the 2026-08-14 self-audit: #31 (feature-template false positive) is fixed in 641784c, Components now scores 16/16. #32 (README media false negative/badge marker) has an open PR (#33, floze-the-genius) — CI green, well-tested, but currently has a merge conflict against main (trivial, adjacent test-fixture insertions in tests/run.sh; will resolve on rebase). Not merged yet, flagging for review.
Nothing new to report against #15's harness gate — Codex CLI and Gemini CLI are still not installed on this machine, and Cursor is still GUI-only, so it's still not live-trigger-testable from here. No change since the 2026-08-14 comment.