poker, bots and stress testing

So, I have found my new challenge :-)
It started almost incongruously (these things have a habit of always starting this small and then getting a life of their own and then beginning to consume the person body, soul and everything). I had already tested 12 bots (those fake players who sit on poker tables and try to behave like real players) sit on some tables and play for some hours. That had given me some serious headache on trying to reconcile the rake with the win amount and with the total loss of real chips of users. But this time the challenge was to make 200 bots sit on tables and play. We started small with a total pool of 200 players with around 50 players sitting on the tables at any point of time. And the results were outrageous. The total loss amount didn't match with the rake and neither did it match with the total loss of chips. It made me crazy. Why would it work with 12 bots and not with 50 bots? The answer of course was not in the numbers, that was a red herring, a fake trail. The real answer was in the way bots were being disconnected and what was happening for these disconnected bots. That led me to some very good observations and some great fixes. Last time around we were able to make 50 bots sit on tables for close to 18 hours when we had to intervene and stop them from playing. That is something, so eat your heart out LK and AC :-)
On other front, I finished the Gaudy Night and it is a good book, a lazy read but with great characters, a great story with fake trails strewn all around and with motives floating around. Really very good read, kept me awake 2 nights in a row trying to finish the book :-)