my simulation: def sim_bios(n, doubleNVRAM=False, D=45): deck = [1]*n + [0]*(D-n) #NV - Pastebin.com
results for at least one 3-of:
andy, no mulligan 0.4975 (agrees with NRDB, cool)
andy, with mulligan 0.7463
bios 0.7108
doubleBios 0.8310 (if the NVRAM stuff triggers on the original draw and the mulligan)
I tried putting together the closed form of this earlier today and got 71% for bios (but then saw the first sim results and thought “I guess I messed something up”).
So, the takeaways are qualitatively similar to @zagzagal’s simulation and @sappidus’s wolfram alpha magic: bios and andy are roughly similar in terms of early consistency. If Bios gets to do NVRAM baloney before drawing the mulligan hand, then she is significantly more consistent.