The query that is used to detect flaky tests on CQ has a bug: https://cs.chromium.org/chromium/infra/appengine/findit/services/flake_detection/flaky_tests.cq_false_rejection.sql?l=301.

The bug is that it assumes all the failed tests in a flaky build are flaky tests, however, this is not necessarily true. For example, in the following scenario:

base_unittests (with patch) failed with t1 and t2, and base_unittests (without patch) failed with t1. In this case, base_unittests (with patch) is deemed as a flaky build because the matching "retry summary" step is red, which is correct, however, t1 shouldn't be considered as a flaky test because it also failed in base_unittests (without patch).

This bug may cause false positives that labels consistent failures as a flaky test, and the correctly solution is to filter out failed tests that also failed in the "without patch" step.