When to Ignore the Solver: Five Spots Where GTO Costs You Money

A solver will tell you how not to lose. It won't tell you how to win the most.

Game-theory-optimal poker is a defensive masterpiece. If played well, the maths are pretty solid. But "can't be beaten" and "wins the most" are not the same thing. Against players you actually sit with, reading the table is where your real profit lives.

Jonathan Little (two-time WPT® champion) put it best. Play a GTO strategy against opponents who aren't brilliant, and you leave a lot of money on the table.

Here's what the solver guarantees. Here’s where following it could cost you. There's one rule that keeps deviating from turning into disaster.

GTO is an equilibrium strategy. Its promise is narrow and specific: play it and no one can beat you. It's made to at least break even against any strategy. And, you profit only when an opponent makes a mistake the equilibrium happens to punish.

The guarantee is that no one can exploit you.

But, notice what's missing. It doesn't say you make the most money.

A GTO strategy is passive by design. Rather than hunting for mistakes, it declines to make any of its own. It lets opponents cost themselves money when they err. Little frames the choice as passive versus active exploitation. Passive means playing the equilibrium and letting errors pay you. Active means deviating from it to attack a leak you've identified. You're taking the bigger profit that comes with the bigger risk.

It's worth being precise about what's even solved. Heads-up limit hold'em was weakly solved back in 2015 by Cepheus. This means that over a long sample, you can't beat it. It does not mean that it wins every hand.

Heads-up no-limit is quite approximated. For example, in 2017, the Libratus bot beat four elite pros by a decisive 14.7 big blinds per 100 hands, and it bluffed to do it. But full-ring, multiway no-limit hold'em is not solved.

The tools you use approximate an equilibrium. They don't hand you a proven answer. In several common situations, the approximation is exactly where things go wrong.

The solver assumes a balanced opponent. An opponent who bluffs the right amount so that you're forced to call the right amount. That assumption is the foundation of every river-defence frequency it gives you.

Most players at low and mid stakes are not balanced. At times, they under-bluff the river. When a passive $1/$2 or $2/$5 regular fires a big bet into you on the end, that action can be valuable. Far more often than the solver's balanced opponent would suggest.

We have the GTO instruction to "call enough to stay unexploitable". Often, it's defending against bluffs the players aren't making.

Against those players, over-fold. Let go of the marginal bluff-catchers the solver wants you to call with. Like on a bad day fishing, there's often nothing to catch.

The mirror image matters too. Against the player who can't stop firing. The one who bluffs every missed draw and then some. You do the opposite and call down light with hands the solver would fold.

Same principle, opposite read. You're adjusting to the bluffing frequency at your table, not the theoretical one.

Standard solver output usually ignores rake. It solves the pure game, as if the house took nothing. This is not the case with real cash games.

Rake changes the math on every close decision.

Let's take a hand that's razor-thin. A profitable call in the solver might become a loser. Why? Once you account for the cut the house takes from the pot you're fighting for.

The same goes for marginal opens and speculative calls. The solver's break-even spots sit a touch on the losing side of the line once rake is in the picture.

In practice, that means playing a bit tighter at the stakes where rake bites hardest. For example, folding the very hands the solver calls close, as with rake this can be unprofitable. The lower the stakes, the more this matters.

Most solver output is chip-EV. It treats every chip as equally valuable and tells you the play that wins the most chips on average. In tournaments, chips are not money, and that difference can flip a "correct" play into a costly one.

On a bubble or at a final table, your tournament equity is governed by the payout structure, not the chip count alone. Busting costs you far more than the chips suggest, because you lose your shot at every pay jump above you. This is the risk premium, and a chip-EV solver doesn't feel it.

It will recommend a thin all-in that's a small chip-EV gain and a real-money catastrophe. It's the kind of flip that's fine in a cash game and disqualifying on a pay jump. Correct ICM play is usually tighter than the raw solve, sometimes a lot.

There are tools that now model ICM. The fix is often to solve the right game rather than to abandon all solvers. But, if you're reading a chip-EV output on a bubble, you're reading the wrong map.

Heads-up, solvers are on firm ground. Add a third and fourth player and the ground gets soft.

Multiway pots are hard to solve. The game tree explodes. Unlike the heads-up case, there can be many equilibria, so there isn't one clean "GTO answer" to memorise. Worse, the heads-up instincts most players have drilled don't completely transfer. Bet sizes. Bluffing frequencies. Defence. All shift when there are more players left to act behind you.

Taking a line that's correct heads-up and applying it four-handed to a limped pot is one of the most common and expensive mistakes strong solver students make. When the pot goes multiway, lean harder on sound fundamentals. Go with caution rather than on a heads-up-derived sim. The tool is least reliable exactly here.

The solver's opponent is a ghost. Unknown, balanced, and giving off no physical information. Your live opponent is nothing like that.

Live low-stakes fields skew in ways you can name in advance. They limp too much. They don't bluff often on the river. They under-defend. Their timing is a dead giveaway along with their posture and their bet sizing. These are all reads, and they're money.

Punishing habitual limpers, value-betting thinner than the solver dares because they call too wide, and over-folding to their rare river aggression. These are edges that pure GTO will never capture. The GTO refuses to assume the imbalance that's sitting right in front of you.

If you have a genuine read (physical or statistical), it can be worth more.

Everything above comes with a single, non-negotiable condition. Deviate only when you have a reliable read, and only against opponents who won't adjust back.

That's the discipline that separates exploitation from spew. The moment you leave the equilibrium to attack a leak, you open a leak of your own. If you have a good, attentive player, they will find it and turn it against you fast.

If your read is wrong, or your opponent notices the adjustment and counters it, you can lose. Sometimes more than you ever stood to win. This is why GTO is the correct default against strong, unknown, or paying-attention opposition.

You don't deviate because deviating is clever. You deviate because a specific opponent is making a specific mistake you can name, and you go back to the baseline the moment that stops being true.

Which is also the answer to why you should study solvers at all, if they're this flawed. You study them as your floor and your defence. The reference point that tells you what balanced looks like, so you can measure how far a given opponent strays from it and how far you can safely lean the other way. The sim is the map you learn before you learn where the map is wrong.

Close it when the table gives you a better reason, and open it again the moment it doesn't.