How do you read a solver output?

Read it in three passes. First the strategy layer, which tells you how often each hand takes each action. Then the EV layer, which tells you how much choosing the wrong action costs. Equity comes last and is rarely worth acting on by itself.

What does EV mean in a solver output?

EV is the expected value of an action in chips, usually shown in big blinds. If raising is worth 2.41bb and calling 2.39bb, the two options are effectively identical and you can simplify. If one action is a full big blind better, that hand is worth drilling until the decision is automatic.

Should you copy solver mixed frequencies exactly?

No. The solver mixes because two actions earn almost the same amount, so any split between them performs the same against a perfect opponent. Copy the pure regions of the range faithfully and pick a simple rule for the mixed band.

Why does a solver fold a hand with more equity than one it calls?

Equity only measures showdown win rate. It ignores position, initiative and whether you can realise that equity across three streets. A hand with 42% equity that can barrel is often worth more than a hand with 48% equity that must check-fold.

What frequency is too small to matter in a solve?

Anything under about 5% is usually convergence noise rather than strategy, and would shrink or vanish with more compute. Round those actions to the nearest pure decision. Above 15% you are looking at a genuine mix worth understanding.

How to Read a Solver Output Without Getting Lost

A solver output stacks three numbers on top of each other for every hand in the range: how often to take each action, what each action is worth, and how much equity the hand holds. Read them in that order — strategy, then EV, then equity — and the grid stops being a wall of colour. Nearly everything that confuses people about solver output comes from reading those layers in the wrong order, or from taking a 3% action seriously.

The three layers in every solver grid

Open any solver — a desktop tool like PioSOLVER, a web-based presolved library, or a trainer built on one — and the display works the same way underneath. You pick a node in the game tree and get a 13×13 grid of starting hands, or a list of postflop combos, with each cell carrying three layers.

Layer	What it answers	When to read it
Strategy	How often this hand takes each action	Always, and first
EV	What each action is worth in big blinds	When two actions look close
Equity	How often the hand wins at showdown	Almost never on its own

The colours you see by default are the strategy layer. Green for raise, blue for call, grey or red for fold — the palette varies between tools, the meaning does not. A cell that is 70% green and 30% blue means the solver raises that hand seven times in ten and calls the rest.

The other two layers are a second click away, and most players never make it past the colours. If the machinery underneath is still fuzzy, start with what a poker solver actually does before you try to interpret what it produces.

Read the strategy layer first

Strategy answers the only question you can act on at the table: what do I do with this hand? Begin there every time, and look for three things in this order.

The pure regions. Find the hands that take one action 100% of the time. These are the backbone of the range and the parts worth memorising. AA raising from every seat is not interesting; the pure fold boundary at the bottom of a UTG opening range is.

The boundary. Somewhere between pure raise and pure fold sits a band of hands that mix. That band is the real output of the solve — it marks where two options were worth almost exactly the same.

The shape. Read the grid as a picture rather than 169 separate cells. Suited hands extend further than their offsuit twins. Pairs form a block. Ace-x suited runs deeper than ace-x offsuit. Those shapes transfer between spots. Individual cells do not.

EV is the second layer, not the first

EV tells you what each action is worth in chips, normally in big blinds. It answers a different question: how much does getting this hand wrong actually cost me?

That is the number that turns a wall of frequencies into a study plan.

If two actions sit within about 0.05bb of each other, the choice is nearly free. If one is a full big blind better, that hand is worth drilling.

A hand the solver raises 55% and calls 45% might carry a raise EV of 2.41bb against a call EV of 2.39bb. The mix looks dramatic and the cost of always raising is two hundredths of a big blind. Simplify it and move on.

A hand the solver calls for 1.8bb where folding is worth nothing is the opposite case. There is a full big blind sitting on the table every time you fold it. Sort your review by EV gap rather than by frequency — the largest gaps are the cheapest improvements you will find.

Anything under 5% is noise

Solvers converge, they do not finish. Run a solve to a low exploitability threshold and you will still see actions taken 2%, 3% or 4% of the time that would shrink or disappear with another hour of compute.

Treat those as rounding errors:

Under 5%. Ignore it and round to the nearest pure action.
Between 5% and 15%. Worth knowing the hand can do it, not worth doing at the table.
Above 15%. A genuine mix that deserves a rule.

The practical test is whether deleting the action would change anything you do. If you would never 4-bet 87s from the SB three times in a hundred, and could not execute it reliably if you tried, that cell is information rather than instruction. Strip it out and the range you are left with is the one you can actually play.

Static charts can't adapt to your opponents

Sharkling's neural solver does — it re-solves the spot for the table you're actually sitting at, then drills you on it.

Try Sharkling free

Equity is the layer that misleads

Equity is how often a hand wins at showdown against the opponent's range. It is displayed prominently and it is the least actionable of the three layers.

Equity ignores position, ignores who holds the betting initiative, and ignores whether you can realise it. A hand with 48% equity that has to check-fold half the time is worth less than a hand with 42% equity that can barrel three streets.

Run it on a concrete pair of hands. A5s and K9o hold broadly similar raw equity against a BTN opening range, yet the solver defends A5s from the BB far more readily. The suited ace makes nut flushes, blocks the strongest part of the opponent's range, and still has something to do on later streets. K9o does none of that.

That is why solvers fold hands that look like coin flips and call hands that look behind. The solve prices the entire future of the hand, not the showdown. When equity and strategy disagree, the strategy layer is right.

Why the same hand does two different things

The first time you see AJo raising roughly two-thirds of the time and folding the rest, the instinct is to hunt for the tiebreaker. There is none. The solver is indifferent between the two actions, so any split earns the same against a perfect opponent.

In practice that means you do not have to reproduce the ratio. You have to keep the pure regions pure and avoid being predictable at the boundary. The reasoning behind mixed strategies in solver output matters far more than the exact percentages, because those percentages move the moment the inputs move.

The output is only as good as the tree

Every number is conditional on the tree you built. Change the bet sizes, the stack depth, the rake or the starting ranges, and the output changes with them.

Rake. A rake-free solve opens and calls wider than any real game justifies. If your tool lets you set rake, set it.
Bet sizings. Give the solver one size and it uses that size perfectly. Give it three and the strategy for every hand shifts.
Ranges at the node. A postflop solve inherits whatever preflop ranges you fed it. Wrong ranges in, confident nonsense out.

When you want to know how a specific opponent tendency changes the answer — someone who never folds to a river bet, say — locking a node in the tree fixes one player's strategy and lets you read the exploit straight off the other side.

A reading order you can repeat

Run this sequence every time you open a node.

Look at the shape of the strategy grid before any individual cell.
Mark the pure regions. Those become rules you can take to the table.
Find the boundary and note which hands sit on it.
Switch to EV and measure the gaps at that boundary.
Round away anything under 5%.
Write one sentence describing the range in plain words.

That last step is what makes the session stick. A grid you stared at is gone by tomorrow; a sentence such as UTG opens about 16% here, all pairs, broadways and suited aces down to A5s survives.

For preflop you rarely need to run a solve at all. A library of solver-generated charts by position already holds the converged output for every seat, and reading one is the same three-layer process with the arithmetic finished for you.