Hi,
I can't seem to reproduce despite running them for a while continuously. Would you be able to share the contents of the Proced buffer before and after the refinement in a flakey test run in addition to the PID refined on?
It might also be worth testing proced-refine works as expected for you when refining by PID (in case the flakiness is hinting at some kind of legitimate bug)?
Thanks, Laurence