A solver answers exactly one question: how should two perfect players, with these exact ranges, these exact stacks, these exact bet sizes available, divide this exact pot? It is a very good answer to that question. The trouble is that nobody at your table has those ranges, your site takes a cut the solver never modeled, and the bet sizes you’re allowed to use are not limited to the four the solution was run with. Strip the question away and keep only the answer, and you’re memorizing the output of a model whose assumptions you never saw.
This isn’t the usual anti-theory complaint. Equilibrium study made a generation of players better, and the top end of the game is unrecognizable from where it was fifteen years ago because of it. But there’s a specific, repeatable way it goes wrong. A player runs a sim, sees “check 71%,” writes “check the turn here” in a notebook, and walks away having learned a frequency instead of a reason. Six months later he’s checking turns in spots that only superficially resemble the original, against opponents who would have paid him off, and he can’t articulate why he’s doing it beyond the fact that the machine said so.
What follows is an attempt to be precise about where the leak is, because “use solvers wisely” is advice nobody can act on.
What the machine is actually answering
A solver takes a starting point you define. You feed it two ranges of hands, a board, a stack size, a pot size, and a menu of legal bet sizes. Then it grinds toward a strategy pair where neither player can improve by changing anything. Every one of those inputs is a decision you made, and every one of them is a place the output can quietly become about a game that isn’t yours.
The ranges are the biggest offender. If you type in a button opening range of 48% of hands and your actual opponent opens 22%, you have solved a different game. The board that “massively favors the preflop raiser” in your sim may be close to neutral against a range that’s all broadways and pairs. Nothing in the output tells you this. The solution looks exactly as confident either way.
Bet sizing menus do the same thing more subtly. Give the solver only half-pot and pot, and it will tell you a beautiful story about those two sizes. It will never tell you that a quarter-pot stab would print against this particular field, because you never offered it that option. The strategy is optimal within the box you drew.
The 68/32 problem, and why memorizing a mix teaches you nothing
Open almost any postflop node and you’ll find mixed actions: call 68 percent, raise 32 percent. New students treat that split as a commandment and agonize over randomizing properly. It’s close to the least useful number on the screen.
Here’s why. A solver mixes between two actions when their expected values are essentially equal. That’s what mixing means. If calling is worth 4.10 big blinds and raising is worth 4.05, the solver will split, and your choice between them costs you five hundredths of a big blind. At a dollar a big blind, that’s a nickel. Meanwhile, choosing the wrong one against a real opponent who folds too much to raises can cost you two or three big blinds in the same spot.
The valuable information in a mixed node is not the ratio. It’s the fact that the two actions are nearly tied, which tells you the hand sits on a knife’s edge, and that any read at all should decide it. A player who learns “68/32” learns a number. A player who learns “these are worth the same against a perfect opponent, so the decision belongs to my read” learns something he can use at a table.
A river bluff-catcher, worked all the way through
Say the pot on the river is $100 and your opponent bets $50. You hold a hand that beats every bluff and loses to every value hand.
You’re risking $50 to win $150, so you need to be good 25% of the time. At equilibrium, a half-pot bet is balanced when bluffs make up a quarter of the betting range. Three value hands for every bluff. If he arrives with nine value combinations, equilibrium says three bluffs, and you’d be right exactly 25% of the time. Calling breaks even, which is what balanced means.
Now take the same spot against a live $1/$2 regular who bets the river only with hands he thinks are good. Give him the same nine value combos and two bluffs. His betting range is eleven hands, you beat two of them, and you’re good 18% of the time. Run the arithmetic: 0.18 times $150 is $27, and 0.82 times $50 is $41. Calling loses about $14 a go. The solver calls this spot a break-even call and the correct play is a clear fold, and the thing that moved them apart was a single combination of bluffs.
Flip it. Against a player who turns every missed draw into a bet, give him nine value hands and twelve bluffs. Now you win 57%. Calling is worth about $64. Your bluff-catcher should be calling with everything that beats a busted draw, far past any frequency the solution endorses. The money in poker lives in that gap between your opponent and the equilibrium, not in the gap between you and it.
Rake doesn’t exist in the solution, and it exists everywhere on your screen
Most sims are run rake-free. Real pots are not. A typical online cash game takes around 5% with a cap of a few dollars, and that cap is what makes the distortion uneven.
A $300 pot paying $3 loses 1% to the house. A $40 pot paying $2 loses 5%. So the smallest pots, the ones that preflop marginal calls and blind defenses are built to win, are taxed hardest. A suited connector call that the solver grades at plus a tenth of a big blind can be a losing call at a real table, because the pots it wins are small pots and the rake eats the edge whole. This is also why solver-approved wide defending ranges go so badly for players who import them from a sim into a high-rake environment, and why game selection decides more of a small-stakes result than any turn frequency does.
The fix isn’t to ignore marginal spots. It’s to treat every near-zero-EV action in an unraked sim as a fold until proven otherwise.
Why a correct call can still be a mistake in your hands
A solution is a complete strategy, not a single decision. When it defends a weak suited hand from the big blind, that defend is only worth what it’s worth because the solver goes on to play every turn and river perfectly afterward: floating correctly, raising the right rivers, folding the right rivers.
You will not do that. Nobody does. Equity realization, meaning how much of a hand’s raw chance of winning you actually convert into chips, depends on how well you play the streets ahead. If your postflop execution with 95 suited out of position captures half of what the solution captures, the defend becomes a leak. The honest version of the question isn’t “is this call positive in theory.” It’s “is this call positive given how I play the rest of this hand.” Those have different answers, and only you can supply the second one.
Making the output explain itself
Three habits turn sims from memorization into understanding.
First, lock a node. Most tools let you force one player into a fixed behavior, then re-solve. Tell the solver your opponent folds 70% to a turn bet instead of 55% and watch which bluffs appear. The change in strategy is the lesson, not the strategy.
Second, run the same spot across a family of boards. Ace-high rainbow, ace-high two-tone, king-high connected, low paired. The frequencies will differ. The reasons, which hand classes want to build the pot and which want to keep it small, will repeat. Those reasons travel to the table. Numbers don’t.
Third, learn the simplified version first. Play Optimal Poker by Andrew Brokos works through toy games small enough to solve on paper, which is the fastest way to see why equilibrium behaves the way it does before you ever open a real sim.
A solver can’t tell you whether this opponent is scared, tilted, short of his buy-in, or simply bad. It isn’t supposed to. It gives you the baseline against which everything those people do is a mistake. Knowing the baseline and nothing else just means you’re playing a flawless game against an opponent who isn’t there.