Every GTO library claims its solutions are accurate. The useful question is accurate compared to what, and how much any remaining error matters for the way you actually study. This guide explains how solution accuracy is measured, where the error comes from, and which kind of inaccuracy you should care about.
How accuracy is measured: exploitability
Solvers don't find the exact equilibrium. They iterate towards it and stop when the strategy is close enough. "Close enough" is measured by exploitability, also called Nash distance: how much a perfect opponent could win against the solution by playing the best possible counter-strategy.
It is usually expressed as a percentage of the pot. A solution with 0.3% exploitability means a perfect, fully informed opponent could gain at most 0.3% of the starting pot per hand against it. Library solutions are typically run to a fraction of a percent; anything a human could detect at the table is far larger.
Lower is better, but with steeply diminishing returns. Going from 1% to 0.3% mostly moves frequencies on close, mixed hands. It rarely changes what a hand does.
Where the error actually comes from
Convergence is the smallest source of error in most solutions. The larger ones come from the assumptions baked into the game the solver was asked to solve.
Bet-size abstraction
A solver can only use the sizes it is given. If the tree allows a 33% and a 75% pot c-bet, the strategy is optimal within that menu. Real equilibrium play might prefer 50%. Studies of simplified trees show that reasonable size menus lose very little EV, but the solution is always "GTO for this tree", not GTO for poker.
Input ranges
Postflop solutions start from preflop ranges. If those ranges are off, every street after inherits the error. This is the most common reason two libraries disagree about the same flop.
Rake, antes and payouts
Rake makes thin calls and flats less profitable, so a rake-free cash solution plays looser than a rake-aware one. Tournament spots need antes and, near the money, ICM rather than chip EV. A solution with the wrong economic model can be perfectly converged and still wrong for your game.
Multiway pots
Most postflop work is heads-up because multiway trees explode in size. Multiway solutions exist but rely on heavier simplification, so treat them as directional rather than exact.
Which inaccuracy matters for you
| Source of error | Typical impact | Should you worry? |
|---|---|---|
| Convergence (0.2–0.5% pot) | Small frequency shifts on mixed hands | No |
| Bet-size abstraction | Small EV loss, different "look" | Rarely |
| Wrong preflop ranges | Distorted postflop strategy | Yes |
| Wrong format (rake, ICM, stack) | Systematically wrong ranges | Yes |
| Your execution at the table | Often many times the above | Most of all |
The last row is the point. The difference between a 0.3% and a 0.5% solution is invisible next to the cost of folding a hand that should always defend. Pick a solution built for your format, stack depth and rake, then spend your energy on executing it.
How to read a solution without over-trusting it
- Look at EV, not just frequency. If raise and call are within a tiny fraction of a big blind of each other, either is fine.
- Trust the pure regions. Hands that always take one action are robust across solver settings. Mixed hands are where settings move the answer.
- Sanity-check the inputs. Stack depth, positions, sizes, rake or ICM: confirm they match the game you play before memorising anything.
- Prefer simple strategies you can execute over complex ones you can't. A slightly simplified strategy played consistently beats a perfect one misremembered.
Turning accurate solutions into accurate play
Accuracy on the screen is worth nothing if it doesn't reach the table. That gap is closed by drilling: being dealt the spot, deciding, and getting graded until the answer is automatic. Our guide on using a GTO poker trainer covers how to structure that, and finding leaks in your own hands shows where your execution drifts from the strategy.
New to the concept? Start with what GTO poker is.
Frequently asked questions
What is a good exploitability for a poker solution?
Library solutions are usually run to a fraction of a percent of the pot. Below that, further convergence mostly moves frequencies on close, mixed hands and rarely changes what a hand does.
Why do two solvers give different answers for the same spot?
Usually because of different inputs: starting ranges, allowed bet sizes, rake, antes or payout model. Convergence differences are typically much smaller than input differences.
Are simplified GTO strategies worse?
Slightly, in theory. In practice a simplified strategy with fewer sizes loses very little EV and is far easier to execute consistently, which usually makes it more profitable at the table.