1) brownout
2) power down
3) power up
4) badly designed uC i.e. reset is clocked, not DC

The above is pure speculation, have no proof or whatever.

However, when it is proven not by actual measurements, but by all senior members here, (with the exception of Richard), having had the problem disappear by a solid powerup/powerdown/brownout detector resetting the uC, then that should be good enough.

I can not say this for sure, but it seems that liking Dallas uCs and not believeing supervisors will fix the problem goes together (see 4) above)

Erik