I worked for a defense contractor many years ago, doing troubleshooting
thru final test on avionics switching power supplies. Part of the
testing was hot/cold cycling under changing load with measurements taken
at every cycle.

The last unit of the run kept intermittently failing the hot cycle with
lousy regulation on load changes. Management wanted it "out the door" ASAP.

Other technicians had pretty much "shotgunned" every part on the board
before it was handed to me.

I spent days trying to duplicate the failure on the bench by standard
troubleshooting.

Finally I went to lifting "possible" parts from the board and spot
heating every thing. I finally stumbled on a 22v zener that showed
minute leakage (about 10 ua) at about 100F. I hooked up a lab supply on
it and curved it out and heated it some more . At 160F it showed a
partial "knee" at 17 volts with about 1ma leakage, then dropped off
until it hit the rated 22v knee. Since it was on one input of a
differential amp used as a reference, this explained the failure. The
1 ma draw was enough to give an erroneous output on the diffamp.

I drew out 40 of the same zener out of stores and found that 6 of them
had the same type of failure. There had to be more of them installed in
supplies that had already shipped, and there was no way of tracking
which units got one of these knock-kneed zeners.

The only saving grace was that the customer's rep said that the units
would never see that kind of condition onboard the aircraft. The temp
cycling conditions were such that they figured that there was plenty of
margin in real world operating conditions.

What's *YOUR* tale about weird component failures?