[jwhat]

Hi Vishnu,

np, I was thinking of complicated possibilities, when it was a simple thing.

Typical programmer mindset ;-)

Why not put up on github and then people could put MacOS and Windows UI on it...

Cheers from Oz,

jwhat/John.

[vishnu]

I've got it on my todo list to port it to WxWidgets for cross-platform use...

[vishnu]

Well I finished putting all the simulation code back in but it needs to run vastly more times than I thought it would in order to generate the exact same numbers Braun got. For example in 180 runs (of 100,000 hands each) it duplicates most of Braun's numbers but there are still a bunch that are one or two off. For example Braun predicted that a player would be dealt two tens versus a dealer's 10 up 2534 times, in the 180 simulation pass that I just did I got 2533, but I think it's safe to conclude that Braun's simulations from 60 years ago are correct. So at this point the question becomes, if you're Braun's supervisor at IBM in 1970, what do you think about him using a million dollar computer to run card game simulations for hundreds, if not thousands of hours? Plus, did he do all the coding on his own time, just staying late after everyone else has gone home? Because I'm pretty sure he didn't have one of those computers at his house...

[robespierre]

A 1960s era IBM S/360 had a pre-IEEE754 floating point capability and may accumulate numerical errors differently from modern machines. (If floating point numbers are involved at all in this program, which I didn't check).

[vishnu]

Nope, no floating point, all 340 possible card combinations are unsigned short int. I just ran the simulation again and got 2535 for 10 10 vs. dealer's 10 up, which is Braun's number. Floating point is definitely complicated, a prime example being the Pentium FDIV bug.