How about using a silicon serial number on each board as a seed ?
Thought about that one, but in a number of instances those will be identical in the last 6 bits. and thus seeding rand()with the same number. I am in a situation where I can not even take 1 in 1000 needs to replace one (which one) of the units.

I see several possibilities, but can not prove either to work and if this goes out and occasionally misses startup I am out of a job.

I am trying to avoid the amplified diode noise setup purely for space reasons. The board size stem from simpler days and need to be the same.

Erik