Why per-repository CI misses it

The web team ships account sharing and its build is green. The Android build is green too, because nothing on Android tests account sharing yet. Each repository passes. The product doesn't, and nobody finds out until a customer asks why the button isn't in the app.

How it works

  1. Each platform is a surface. A project lists its surfaces (for example web, iOS, Android and API), and each can point at its own repository.
  2. Each requirement says which surfaces it needs. Account sharing is required on web and Android, while barcode scanning might be mobile-only.
  3. Tests in each repository carry the same marker.
    // web:      fp:share_account@web
    // Android:  fp:share_account@android:e2e
  4. Every repository's CI reports to the same matrix. One row per requirement, one column per surface: passing on web, missing on Android. That gap is visible to everyone, without opening either codebase.

Closing the gap

The missing cell is also a work order. An agent working in the Android repository runs fp list --gaps, reads the requirement and the web evidence, builds the feature, and tags its test. When Android CI reports a pass, the cell turns green. At the covered enforcement level, fp check fails the build while a required cell has no evidence.

Limitations

  • Each repository has to run fp ci-report in CI; a platform that never reports stays missing.
  • Which platforms a requirement needs is a product decision. Leave a platform off the requirement and it is out of scope there, not a gap.
  • FeatureParity compares evidence, not behaviour. Two platforms passing their own tests can still differ in ways no test checks.

See also: verifying agent-written code.