[Copernicus]

I'm very familiar with the Alberta work. In fact a lot of work I have done paralleled theirs, and actually took the Poki concept several steps further when they sidetracked into the heads-up model. Aaron Davidson was most helpful in discussing a conceptual framework for that next generation.

Unfortunately I don't have the computer skills or funding to turn the work into a "bot" (or more to the point an automated but off-line "advisor", since bots are violations of the TOS of almost all sites). The limitations are not in processing since the results arent necessarily optimal, but rather in I/O of the real time situation. Tested against real hand histories after the fact, however, it learns and predicts other players actions/hands very well. Its impossible to turn that into a theoretical earnings rate, however, since the opponents actions might have been different based on the actions of the "advisor". It is also a monumental data mining task that would benefit from a "team" of players contributing hand histories to be most effective.

I agree with your definition of "optimal", and that is why I commented that there is no optimal game theory solution, since there will always be a counter-strategy for any given fixed strategy. That is why the best poker players will eventually break very good but rigid (although complex) players. (I distinguish game theory solutions (fixed) from AI (learning/genetic) solutions.)