820519cb0c
Run 2 put the inducer on from flight start and froze ~96s in, against 670s clean with it off -- n=2, contrast sharp, confounder (elapsed mission time) still untouched. Withdraws last iteration's 'T74/T75 move off a semaphore onto an event' as the signature to chase: it does not reproduce. In run 2 they are on XEvent while HEALTHY and stay there. The healthy state varies between instants, so a one-sample-per-state diff cannot separate a freeze transition from ordinary variation -- I read a difference of samples as a difference of states. Reproduces across both: T68 and T69 go from not-waiting to waiting, T69 on a semaphore and T68 on an event both times. And 21 of 24 threads unchanged in run 2 (17 of 24 in run 1), so 'not a whole-emulator stall' now has two independent captures behind it. Next: repeat the capture several times within one healthy run to establish which thread states are stable before reading any frozen diff.