Why Solvers Mix Strategies Instead of Picking One

When a solver shows a hand as 62% raise and 38% call, it is not hedging or expressing doubt. Both lines earn exactly the same EV against a balanced opponent, and the split exists to stop that opponent from ever holding a profitable counter-strategy. Mixing is a property of the equilibrium itself, not a difficulty setting bolted onto your study. Once you understand what the percentages are for, most of them stop mattering to you.

A mix is a tie, not a hedge

A solver reports, for every hand in a range, how often it takes each available action. Most hands come back pure: AA opens 100% from UTG, 72o folds 100% from everywhere. Those are the easy ones.

A split appears only when two actions produce the same expected value. If raising A5s were worth 0.3bb more than calling, the solver would raise it every single time — there is no reason to leave value on the table for the sake of variety.

The way the algorithm works makes this obvious. It plays the game against itself millions of times, shifting weight towards whichever action performed better on the last pass. Weight stops moving when the actions stop differing. A 62/38 split is the algorithm having found a spot where it can no longer improve either way.

The tie exists to make your opponent indifferent

Your own EV is not what determines the frequency. Your opponent's is.

Think about a river spot where you bet with value hands and bluffs. If you bluff too often, calling becomes profitable for your opponent and they call everything. If you bluff too rarely, folding becomes profitable and they fold everything. Between those extremes sits one ratio where calling and folding earn them exactly the same amount — and at that ratio, nothing they do gains against you.

A hand mixes at whatever frequency makes your opponent's decision break even. The split protects your strategy, not that individual hand.

This is the indifference principle doing its work, and it is why the frequencies look so arbitrary. They are solutions to a constraint on someone else's choices, which is why they rarely land on round numbers.

The defensive side of the same idea is minimum defence frequency: MDF = pot/(pot + bet). Against a half-pot bet you continue with about 67% of your range, against a pot-sized bet about 50%. Which hands you use to make up that share is where mixing does its work.

What mixing actually buys you

Take the button opening range. In 6-max, BTN opens roughly 40-45% of hands and UTG roughly 15-17%. Suppose the button decided to simplify by opening only its strongest 40% purely, folding everything else with no overlap.

The big blind can now read the range off a chart. Every button open means the same set of hands, so the blind's 3-bets, calls and flop defence can be tuned against one fixed target. Real mixed preflop opening ranges blur that edge: some hands appear in the opening range only part of the time, so the blind can never be certain which version of the range they face.

The gain is small per hand and relentless over volume. That is the shape of most GTO edges.

Spot	Typical solver behaviour	Cost of picking one side
Premium hands, any position	Pure raise	None, it is already pure
Marginal opens on the BTN or CO	Mixes open and fold	Negligible
Suited aces facing a 3-bet	Mixes 4-bet and call	Small
Flop continuation bets on dry boards	Mixes bet and check	Moderate
River bluff candidates	Mixes bet and check	Large

The pattern in that column is worth internalising. Mixes matter most on the river, where the bluff-to-value ratio is the only thing making your bets work, and least preflop, where the ranges are wide and the alternatives are close in value anyway.

You do not need to randomise at the table

The honest version of this: nobody at low or mid stakes is measuring how often you open KTo from the hijack. Against opponents who are not tracking you, a mixed hand played the same way every time costs a fraction of a big blind per hand, because both lines earned the same EV to begin with.

What actually gains is picking the side that beats the pool you are in.

• The pool folds too often to 3-bets. Take the aggressive side of every close preflop mix.
• The pool calls too much on the river. Take the value side and drop the bluff half entirely.
• The pool never folds a flopped pair. Bet the value mixes, check the bluff mixes.

Randomising is what you do when you have no read. Against a field with obvious tendencies, choosing on purpose beats a coin flip every time.

Where mixing genuinely matters

Two situations change the calculation.

Against opponents who study you. In a regular game with a small player pool, taking the same line with the same hand every time is a tell that gets shared. Here the mix is real defence, and a watch or a card suit works fine as a randomiser.

In spots where the whole line depends on the ratio. River bluffs are the clearest case. If you bet $60 into a $100 pot, your bluffs need to work about 37.5% of the time, and the ratio of bluffs to value in that betting range is what determines whether your opponent can profitably call. Collapse it to pure — always bluffing your missed draws, or never — and the bet stops functioning as a bet.

Turning mixes into decisions you can make

The practical approach is to round the frequencies before you memorise anything.

1. Anything above 80% becomes always. The remaining 20% is worth less than the memory it consumes.
2. Anything below 20% becomes never. Small-frequency actions are the last thing to add, not the first.
3. Genuine coin flips get a rule. Pick the side that beats your pool and apply it consistently until the pool changes.
4. Keep the mix only where the ratio is the point. River betting ranges, and 3-bet defence against opponents who fight back.

This is how a simplified GTO strategy is built: not by ignoring the solver, but by discarding the parts of its output that are worth less than the errors you make from trying to remember them.

The habit that makes all of this quicker is checking the EV column alongside the frequency column. Two actions at 50/50 with identical EVs is a free choice; two actions at 50/50 where one is worth 0.2bb more is the solver's accuracy limit showing, and you should simply take the better one. That distinction is the single most useful skill in reading solver outputs, and it turns a wall of coloured percentages into a handful of decisions you can actually carry to the table.