Can see how that'd come across as a false dichotomy; not my intent or mental model. What I was thinking and shorthanded poorly was - if we said "you need to run CI on all dependents and confirm no regression", that makes utils the single highest-toil place in the ecosystem to get code merged. You'd have to do the A/B triage of your run to a control for every project that consumes it. Definitely not "block on green" - I glossed over that poorly.
There's also an opportunity for us to automate LLM-based triage of flaky or failing tests to help burn those down, potentially giving us a state where we *actually* had green CI ecosystem wide for the first time in over a decade. That'd change the calculus here quite a bit and just make utils merges long / high compute but not a flaky CI nightmare. On Mon, Aug 24, 2026, at 5:36 AM, Mick wrote: > > > 1. full CI pass on everything puts a dependency on us having > > consistently green CI or utils becomes the single worst place to try > > and get something merged in the ecosystem. We haven't had consistent > > green CI in the 12 years I've been around; I don't see how we're going > > to start now. This would probably make utils DOA. > > > Nit: this is presenting a false dichotomy, imho. NOT green means > harder/headache but not "not usable/valuable". Green is not a dependency, > and I don't think that should be presented as a reason to do less testing. > > Moving towards (3) should instead be based on what downstream testing do we > need. If we agree that some set of unit, property and fuzz testing puts us > in a safe place (and every downstream agrees) that should be enough to > warrant the minimalisation fo what CI to run. >
