So we can draw a line between WinFE 3.x and both WinFE 4.x and 5.x.
I have some doubts about calling a Windows PE built from 8.1 sources a WinPE 5.x or more properly a WinPE 4.1, however.
To go back to the "probability trap" I made earlier it is time to explain what it was.
The calculation made to reduce the number of possibilities from 1:4294967295 to 1:65535 is completely m00t , IF (but unfortunately it is not the case ) all signatures would actually be of the type ABAB or even "only" of the type ABCB it would derive that a a signature manually made of the type ABCD (like 01020304 or 0C0D000B or A00CD2FE) will have 0 (zero) probabilities of finding a collision as all automated disk signature would have a"pattern" that our manually made signature would avoid.
The fact that these patterns, though not compulsory, have been observed in a large enough number of occasions effectively decreases anyway the probabilities of a collision i.e. the 4294967295 is a number estimated (largely) by defect. .