Found during the PR #453 review. A CI runner's simulator vanished mid-prepareTestResults.sh (Unable to find a device matching the provided destination specifier), the script's || true swallowed the failure, and the resulting stub .xcresult (zero testNodes; neither backend can render it) passed the fixture verify step, which checks only Info.plist presence. The test job then failed downstream with a confusing error instead of failing fast at generation.

Fix: harden the verify step (shared by test.yml and toolchain-drift.yml via the fixture cache path) to assert each bundle actually contains test data — e.g. xcrun xcresulttool get test-results summary reporting a non-zero totalTestCount, per bundle. That converts a runner-infrastructure flake into an immediate, attributable failure and keeps a stub from ever being SAVED to the fixture cache (the near-miss scenario: a stub cached under a valid key would poison every subsequent hit until the key rotates).

Two smaller notes from the same review, fine to fold in here or fix opportunistically: scope the 'no other fixture test produces Expected Failure' doc comment in the sample app to TestResults.xcresult (RetryTests.testInUnknownState already records one in RetryResults), and consider pinning renderer scale=1 for the programmatic PNG attachment so it is truly 8×8.

Refs #453 (review), #412/#436 (cache), #392/#442 (drift workflow).