How we compute, and how we check
Every answer on Exploit It comes from our own solver, not from a chart copied off the internet. Here is how it computes, what we checked against independent calculations, what we still measure, and what it does not do.
By the numbers
Preflop: a full game tree
For every table size and stack depth, the solver builds the whole tree: open, limp, 3-bet, 4-bet, all-in, call and fold, at every position. It solves it with CFR+, the algorithm behind modern solvers, over 300 to 1,500 iterations. Every hand gets a frequency and an EV for every action.
The tree does not replay the hand after the flop. When the preflop action closes, each hand gets its share of the pot from its equity, adjusted by its equity realization: out of position, a hand realizes less than its raw equity. That adjustment is measured on our flop solver, scenario by scenario.
One size per decision: open to 2 bb (2.2 bb at deeper stacks), then one 3-bet and one 4-bet size depending on position.
Checked against independent calculations
Against Nash push/fold
At 8 bb, the full tree is free to open, limp or go all-in. Nash push/fold, computed separately with a different algorithm, only knows all-in or fold. If the tree is right, both should play almost the same hands. They do:
| First in, 8 bb | Full tree: hands played | Nash push/fold: hands shoved |
|---|---|---|
| HJ | 30.1% | 29.1% |
| CO | 34.2% | 34.2% |
| BTN | 43.5% | 43.9% |
| SB | 77.1% | 74.1% |
6-max, chip-EV, same ante on both sides (0.75 bb). The SB plays a bit more in the tree, which also lets it limp or min-raise.
Push/fold itself
On October 13, we recomputed push/fold with a second program, written separately, using a different hand evaluator (the one from postflop-solver, an open-source solver). That check found a bug: posted blinds were counted wrong at showdown, so all-in ranges were several points too wide. Fixed the same day. Since then both calculations agree within one point: heads-up at 10 bb with no ante, the SB shoves 58.1% of hands and the BB calls 36.1%.
The hand evaluator
3,000,000 showdowns compared with postflop-solver's evaluator: zero disagreements.
The flop
Our flop solver was compared with postflop-solver on the same spots, with the same ranges and bet sizes: same EVs. Frequencies only differ on mixed hands, which earn the same either way.
Convergence, measured
A solver always stops short of a perfect equilibrium. We measure by how much: for every hand at every decision, what it would gain by playing its best action instead of the displayed strategy, weighted by how often that decision is reached. The total reads in bb per hand dealt, all decisions at the table included. The closer to 0, the closer the tree is to equilibrium.
| Stack | Iterations | 6-max | 9-max |
|---|---|---|---|
| 8 bb | 1,500 | 0.00 | 0.00 |
| 10 bb | 1,500 | 0.00 | 0.01 |
| 12.5 bb | 1,500 | 0.00 | 0.01 |
| 15 bb | 1,500 | 0.01 | 0.01 |
| 17.5 bb | 1,500 | 0.01 | 0.03 |
| 20 bb | 1,500 | 0.02 | 0.05 |
| 25 bb | 1,500 | 0.03 | 0.08 |
| 30 bb | 1,500 | 0.03 | 0.08 |
| 40 bb | 1,500 | 0.01 | 0.01 |
| 50 bb | 1,500 | 0.01 | 0.01 |
| 70 bb | 1,500 | 0.00 | 0.01 |
| 100 bb | 1,500 | 0.00 | 0.01 |
Chip-EV, bb per hand dealt. Since October 22, all these trees are solved over 1,500 iterations: at 300, they were up to 0.46 bb per hand away from equilibrium, mostly from 12 to 25 bb. Measured on 2026-10-22 on the live trees. ICM, limper and other table sizes (2, 3, 4, 5, 7 and 8 players): 0.30 bb at most.
What it means in the quiz
- Your grade is the EV you lose, in bb, not the frequency. An action the solver plays at least 2% of the time is always "perfect": in a mix, both answers are right.
- A spot only gets into the quiz if the decision is close, and only if its answer is within 0.05 bb of the best EV at that node. Otherwise it is never drawn.
- The Solver, the strategy pages and the quiz read the same tree. Before every release, a check makes sure no quiz spot contradicts the Solver grid.
What the solver does not do, or not yet
- One bet size per decision.
- On the flop: heads-up pots, one c-bet size and one raise size; turn and river with fixed sizes.
- Tables from 2 to 9 players; no 10-max table yet.
- Facing a limper, the limping range is assumed: a recreational player's (small pairs, suited aces, connectors, weak broadways). That is not an equilibrium.
- It is an off-table study tool, never for use during a session.
Correction log
- 2026-10-22. The last 12 chip-EV trees solved over 300 iterations (6-max and 9-max, 8 to 70 bb) recomputed over 1,500: their distance to equilibrium is divided by 6 to 20. The quiz answer changes on 6% of their spots. Videos rendered before that date show the old frequencies: we only keep scheduled the ones that announce the same answer, within 10 points, and the carousels were redrawn.
- 2026-10-20. EVs fixed in every tree. An export bug skewed the displayed EVs at deep stacks (the range width used for equity realization), while the strategies themselves were right. Found while preparing this page, recomputed and checked node by node.
- 2026-10-20. Heads-up: the engine made the SB act first after the flop, although it has the button. Fixed before the 2-player tables went live.
- 2026-10-20. The quiz table shows the real pot and each player's real bet (before: "Pot 2.5 bb" on every spot).
- 2026-10-19. The quiz no longer draws a spot whose answer is more than 0.05 bb away from the best EV (44 spots removed so far).
- 2026-10-18. The Solver reads the same tree as the quiz. Before, an older table could contradict it.
- 2026-10-13. Preflop equities recomputed over 1,000,000 boards per matchup (instead of 4,000), then every tree and the quiz.
- 2026-10-13. Push/fold bug fixed: blinds counted wrong at showdown (see above).
Found a mistake?
If an answer looks wrong to you, email contact@exploit-it.com with the link to the spot. We check it, fix it, and log it here.