I'm sorry, but I still don't understand what you are trying to do (with either change), so I don't believe I'm in a position to approve or disapprove. Perhaps Meredith Rawls has a better understanding of what this would mean for our workflows (if she's not too busy)?

I feel like you two disagree on the purpose of ap_verify. Both your views make sense to me... it sounds like John wants "fail fast" to be the ap_verify default to make debugging with it go faster, and Krzysztof is concerned this goes counter to the main purpose of ap_verify as a tool to reproduce runs.

Could we compromise by making "fail fast" an acceptable command-line option for ap_verify (so John can use it for debugging when necessary), but NOT having it be the default (since it's not necessarily desired behavior for checking runs can be reproduced)?

If Meredith thinks that's best, then I have no objections.

Adding it as a commandline option is less trivial than this commit, so it might be a while before I get to it.

Following a discussion on #dm-alert-prod about whether ap_verify still has the same purpose it was originally written for (it does not, and I'm the only one who thought otherwise), I've no further objections to the PR as-is.