GTO vs exploitative poker: MDF, node locking and when to deviate

GTO or exploitative poker: do you have to pick one? No. They answer two different questions. GTO asks what you should do if you know nothing about your opponent. Exploitative play asks what you should do once you know something real.

In this article I put both on the same flop, in the Poker Academy postflop solver. First the equilibrium answer, then the answer against an opponent locked as a calling station. The difference is bigger than most players expect, and so is the risk.

What GTO means

GTO (game theory optimal) is a strategy at Nash equilibrium: neither player can change their strategy alone and win more. A solver finds it by letting two strategies play against each other until neither can improve.

That gives GTO one property nothing else has. In a two-player spot, no opponent can gain anything against it by playing differently. It doesn’t mean GTO wins the most. It wins when the opponent takes actions the equilibrium never takes because they lose EV, like calling with hands that should fold. Against other mistakes, such as mixing two equally good actions at the wrong frequencies, GTO wins nothing extra. That’s the gap exploitative play fills.

One caveat: solver strategies are for heads-up pots. Multiway pots and tournament payouts (ICM) break the clean guarantee, so treat GTO there as the best available baseline, not a proof.

What exploitative poker means

Exploitative poker changes the baseline to punish a specific leak. If a player never bluffs the river, you fold more. If a player never folds, you stop bluffing and bet your value bigger.

The trade-off is simple. Every deviation that wins more against the leak opens a leak of your own. If your read is wrong, or the opponent adjusts, the player you tried to exploit now exploits you.

	GTO	Exploitative
Assumes about the opponent	Nothing	A specific, repeated leak
Against a player with a real leak	Wins from some mistakes, but not the maximum	Wins more
If your read is wrong	Nothing to lose	Loses EV, sometimes a lot
If the opponent adjusts	Can’t be punished	You have to adjust back
Where it fits	Unknown or strong opponents, your default	Known pools and players, with enough hands to trust the read

Correct exploitation of a stable leak earns more in the long run than equilibrium play, not just in the short term. The price is exposure, not variance.

Minimum defense frequency (MDF)

MDF is where most players meet game theory for the first time, and where they misuse it most. It answers one question: how much of your range has to continue against a bet so that a bluff with no equity can’t profit automatically?

A bluff that gives up when called risks the bet to win the pot. It breaks even when you fold α of the time. So if you fold more than α, any two cards bluff at a profit (before counting your raises). Against a half-pot bet, say 5bb into 10bb:

Three limits of MDF:

• It doesn’t pick hands. It’s a share of your whole range. Which hands make up that share depends on showdown value, blockers and what beats what. A solver decides that, not the formula.
• It assumes the bettor’s range at that size can hold any hand. If the bettor’s range for that size is strong, folding more than MDF is correct.
• It ignores rake and the bettor’s equity. A bluff with a draw needs fewer folds than α.

A hand-level example, fixed

The old version of this article used this spot: river K♠︎Q♠︎5♥︎3♣︎2♠︎, your opponent bets half pot, and you hold A♦︎5♦︎. It called A♦︎5♦︎ “middle pair” and said you must call it to meet MDF. Both parts were wrong.

Count the cards first. The board ranks are K, Q, 5, 3, 2, so a pair of fives is third pair with an ace kicker. It’s not a straight: A-2-3-5 needs a 4. It’s not a flush either: A♦︎5♦︎ has no spade, and the board has only three spades.

Then the range part. MDF says 66.7% of your range continues against a half-pot bet. It doesn’t say this hand is in that 66.7%. Whether third pair calls depends on what else you arrive with, and on what the bettor’s range holds on a board where three spades got there. I don’t have a solver run for this exact river, so any answer for this one hand is a heuristic, not GTO.

A real node: the solver folds more than MDF

Here is a spot I do have solved. Cash game, 6-max, 100bb, Normal NL200 GG preflop simulation. HJ opens to 2.5bb and CO calls. The flat is rare: in this simulation CO 3-bets 9.3% of hands against the HJ open and calls only 1.8%. The pot is 2.5 + 2.5 + 0.5 + 1 = 6.5bb and both players have 97.5bb behind.

The flop is 8♣︎8♥︎6♦︎. HJ checks 72.6% of its range. First, what does CO do after the check?

CO bets small with most of its range. The 75% bet is almost never used. Now look at HJ facing that rare 75% bet, 4.9bb into 6.5bb:

MDF says 57.0%. The solver continues with 34.3% and folds 65.7%, well over α = 43.0%. That’s not the solver being exploitable. My reading, which is a heuristic: CO bets this size with only 1.8% of its range, so what arrives here is not any two cards, and HJ’s range is the part that already checked. MDF is a warning sign that you might be over-folding. It’s not a rule the equilibrium follows.

Locking a calling station

Now the exploitative side. Poker Academy’s postflop solver has node locking: you fix how one player plays at chosen nodes, and the solver re-solves the rest of the tree against that fixed strategy. You can lock a node by hand or pick a profile: Nit, Bully or Calling Station, each in S, M or L strength. The original GTO solution stays untouched, and the app shows both side by side with the change per action.

This is the same flop, with HJ locked by the Calling Station L profile. First, HJ facing the same 4.9bb bet:

The locked HJ folds 18.8 percentage points less. Here’s how CO’s best response changes one node earlier, after HJ checks:

• Small bets disappear. Bet 25% goes from 65.0% to 10.7%. A small bet exists to get folds from weak hands cheaply, and this player doesn’t fold enough for that.
• The value goes all-in. CO jams 97.5bb into 6.5bb with 71.1% of its range, where GTO jams 0%. If HJ calls with worse, the biggest bet wins the most.
• EV moves. The app’s lock-impact line at this node shows CO +3.31bb and HJ −3.47bb against GTO.

Don’t copy the 71% jam into your game. The Calling Station L lock is an extreme model, and the solver answers an extreme model with an extreme strategy. My guess, not a solver result: a real player folds his worst hands to a 15-times-pot shove, and against that player most of those jams burn money. What carries over is the direction: against a player who calls too much, bluff less, bet value bigger, and stop using sizes designed to get folds.

How to exploit without getting exploited

• Start from the baseline. Know what GTO does in the spot. You can’t see a deviation without knowing what it deviates from.
• Name the leak as a frequency. “He calls too much” is vague. “He folds 30% to flop bets where the solver folds 60%” is something you can lock.
• Get enough hands. One showdown isn’t a read. A tendency seen over hundreds of spots is.
• Lock only what you know, then check the EV. Lock the nodes you have evidence for and let the solver work out the rest. If your EV barely moves against the lock, the leak isn’t worth the risk of deviating.
• Deviate in proportion to your confidence. A mild lock (S) is a smaller bet on your read than an extreme one (L).

Mass data analysis (MDA) is how serious players get those frequencies: large hand databases of a player pool, compared with solver baselines. The pool’s average leak becomes the lock. That’s the same workflow as above, with real numbers instead of a profile.

Two misconceptions

“GTO only works against GTO players.” Backwards. No opponent can gain against GTO in a heads-up spot, and it wins against every player who takes the EV-losing actions the equilibrium avoids. What it doesn’t do is win the maximum from a specific leak.

“Exploitative play doesn’t need theory.” Reads without a baseline are guesses. If you don’t know that the solver folds 65.7% facing a 75% bet in the spot above, you can’t tell whether a player who folds 50% there is a station or a nit.

Cash games and tournaments

The example above is a 100bb cash game. Exploitation is most common there: pools are big, players repeat, and hand databases are large. In tournaments the same logic works, but chips aren’t money. Near the bubble and at final tables, ICM makes calling off your stack more expensive, so both a “station” and a GTO player should call off their stack less often. Lock against what a player does at that stage, not at a cash table.

Wrapping up

Use GTO as the default and exploitative play as a deliberate bet on a read. The solver shows you both: open the spot, look at the equilibrium, lock what you know, and compare the EV before you change your game.

Cheers! 🙂