I'll typically run CI against Debian oldstable, stable and testing; those do change over time (versions of libraries etc) and that change can cause failures. That's desirable and the reason I have a monthly CI run for "finished" projects.

Two main reasons I see frequently:

* Limits on the free plan (runner minutes, timeouts, etc)

* Ignoring hard-to-fix runs: flakey tests are a real problem and a pain to fix