The App Store rejection regression suite — every rejection, forever

Anvil ships a starter library of common rejection categories — paywall ambiguity (3.1.2), privacy policy link presence (5.1.1), in-app purchase restore button (3.1.1), missing permission rationale (5.1.1), and about fifty more. Each one is a YAML spec that describes the precise UI state that failed review and the predicate that proves the fix is in place.

When your team gets a new rejection, you add it as a spec in minutes. The spec is committed to your repo alongside your code — version-controlled, reviewable, diffable.

Start from the rejection email. Paste the reviewer's description verbatim as a block comment at the top of the spec. Then write the minimal steps that reproduce the state they described — usually 3 to 7 verbs. Finish with the assertion that proves the fix is still in place: an element present, a label text, an accessibility trait.

# Rejection: 5.1.1 — submitted 2025-11-04
# "The app's subscription auto-renew disclosure is
# not displayed on the paywall screen."
name: paywall_autorenew_disclosure_present
device: any-ios
app: ${BUNDLE_ID}
steps:
  - verb: launchApp
    params: { bundleId: "${BUNDLE_ID}" }
  - verb: tap
    params: { label: "${PAYWALL_LABEL}" }
  - verb: waitForElement
    params:
      predicate: "label CONTAINS 'auto-renew' OR label CONTAINS 'automatically renews'"
      timeout_ms: 5000
  - verb: captureAccessibilityTree
    save_to: paywall-tree.json

A single-axis rejection spec catches the obvious case. The adversarial matrix crosses it with nine stressful axes — device memory pressure, network weakness, locale with bidi-override characters, incoming call mid-flow, timezone drift, first-launch vs background resume, and more. Pairwise coverage means the matrix stays tractable (under 400 specs) while catching real-world combinations that would otherwise slip.

Every rejection spec gets an adversarial expansion automatically, written to specs/adversarial/<flow>/. You review them once, then the matrix script regenerates them when the axes catalog changes.

Google Play's policy rejections are rarer but equally damaging — especially around Target API level, data safety, and background location. The same YAML format works on Android by changing device: any-ios to device: any-android. The verb surface is identical.

A mature suite holds 50-200 rejection specs across both stores. Anvil customers report the suite pays for itself the first time it catches an in-development regression of a fix from two quarters ago.