Quote:
Originally Posted by AceHighIsGood
It goes evem further than that. The ranges are optimal for his play against his opponents. There are errors on both sides. Solvers assume perfect play from your opponents as well as from you. In other words, his ranges are exploitative of his group of opponents (in 2004)
As you point out, the solver is recommending things that real players don't do. That is happening on both sides of the analysis! There may be exploitative plays that fare better than the solver output against the strategies that real players use.
Good points, and I agree. The second point is most interesting to me around solver outputs compared to the realities of practical opponents. If we can be specific about the types of mistakes an opponent makes we can model the optimal exploitation in a solver.
For example, let's say we have a CO open vs. BB defend situation. Let's say the board comes out, 3
5
7
, a board which favors the BB's range. A GTO solver model will recommend the BB donk this board with about 45% of it's range and that the CO should bet about 70% of its range when checked to. After a CO bet, the BB should check raise with about 30% of it's range. So let's refer to that as optimal play.
Now let's explore a common heuristic exploitation from normal live play, where the BB always checks the flop and the CO always continuation bets. When we force this action in the solver we see that now the BB is check raising 57% of it's range!
We can take this even one step further and look at the scenario where we allow the BB to have a donk range but we force a bet in the CO. Now the solver tells us that optimal play is to never donk and we can conclude that the optimal exploitation of opponents who do not have a check back range on this board is to always check and have a high check raise %.
Ask yourself, are you attacking these types of boards like this against opponents who always continuation bet in this spot? There are countless opportunities to model these types of exploits in your home lab if you have the time and initiative to do so.
Stepping back for a minute to think about the big picture of all this is probably a good idea. Solvers are very interesting in that they provide a view of what is optimal and unbeatable play for all opponent's involved in a hand. That said, even if you could emulate GTO perfectly in live play, it will certainly not win you the most money against imperfect opposition. You can never lose but your win rate will be lower.
To my mind the best way to use GTO knowledge in practical play is to identify errors that we and our opponents are making compared to the GTO model. From there we can infer exploitative strategies which improve our overall win rate. Without an understanding of what perfect play looks like, any exploitative strategies suggested are educated guesses until we fully understand what a solver optimization looks like in that particular non-GTO scenario.