You drop in a hand history. A few seconds later every decision you made has a grade and a cost in big blinds. This post explains what happens in between: where the GTO answer comes from, how the cost is worked out, and what we leave ungraded on purpose.
The short version: preflop is graded against our own 6-max charts, picked for the stack depth and open size in your hand. Heads-up postflop pots are solved on our own GPU solver for the exact board, pot and bet sizes. Anything we have no solution for is labelled not graded. We never fill a gap with a guess.
What Rangebet grades
- Game: 6-max No-Limit Hold'em cash games, with stacks from 30 to 150 big blinds.
- Preflop: every decision you make, in every pot.
- Postflop: the flop, turn and river of every heads-up pot, meaning two players saw the flop.
- After play only: Rangebet reads the files your poker client saves. It never connects to a table and never runs while you play.
Preflop: the right chart for your hand
A preflop chart only means something for the spot it was solved for. A 100bb chart is wrong for a 40bb stack, and a chart built for 2.5bb opens is wrong when the opener made it 3.5bb. So the first job is to pick the right one.
- Positions come from the blinds and the seat order, not from the site's labels. Tables with three to six players, dead buttons and empty seats all map to the right 6-max positions.
- Stack depth: the effective stack (yours against the deepest opponent) snaps to the nearest of 30, 50, 75, 100, 125 or 150bb.
- Open size: the size of the first raise picks the chart's open-size tree: 2, 2.2, 2.5, 3, 3.5 or 5bb.
- The walk: every action in the hand, yours and your opponents', is followed through the chart in order. When someone makes a raise far from every size the chart has, the walk stops there, and later decisions are marked not graded with the reason.
Limped pots are handled too. At 100bb they use a chart where first-in limps are set to how often the player pool actually limps. At other depths the limp chart is an approximation, and the decision says so. Every approximation we make is listed on the decision it affects.
Some chart spots are reached so rarely that the solve never settled on an answer. We check every spot for that and do not grade the ones that failed the check.
How the cost is worked out
The charts carry an EV for every action with every hand. For your decision we take the EV of the best action for your exact hand, subtract the EV of the action you took, and that is what the decision cost you, in big blinds.
Two details matter here:
- Bet sizes. If you raised to a size the chart doesn't use, the EV comes from the nearest size it does use, and the decision says so.
- Off-chart plays. If you took an action the chart never takes with your hand, for example calling a 3-bet with a hand the chart always folds, there is no EV for it. We grade it a mistake and show the cost as unknown. We do not invent a number, and off-chart plays are not added into your bb/100 totals.
Postflop: solved for the hand you played
When a hand reaches the flop with two players, it goes into a queue as soon as the upload finishes, biggest pots first. Our GPU solver then works through it street by street.
- Ranges: each player's range at the flop comes from the preflop chart line the hand actually took.
- The spot: the solver gets the exact board, pot, stacks and the bet sizes used in the hand.
- The flop runs on a fast tier that values the turn with a trained network. On average it lands within about 1% of the pot of the full solve.
- The turn and river are solved exactly.
Each of your postflop decisions is then priced the same way as preflop: the best action's EV minus the EV of what you did. A solve takes about 7 seconds of GPU time per hand on average, so a session's heads-up pots are usually graded within minutes of the upload.
The grade bands
A grade is a label on the EV a decision lost. Preflop uses these lines:
| Grade | EV lost on the decision |
|---|---|
| Best | Under 0.005 bb, with an action the chart takes at least 5% of the time |
| Good | Under 0.05 bb |
| Inaccuracy | 0.05 to 0.2 bb |
| Mistake | 0.2 to 0.75 bb |
| Blunder | 0.75 bb or more |
Postflop uses the same lines in small pots. In bigger pots each line also scales with the pot, so a few tenths of a big blind in a 60bb pot isn't called a mistake.
Mixed strategies are not mistakes
Solvers often split a hand between two or three actions. Queens might 4-bet most of the time and call the rest. When a strategy mixes, the actions it mixes between have close to the same EV, so taking any of them costs next to nothing and grades best or good.
Postflop, solver numbers carry a little noise. An action the solver takes at least 5% of the time that comes out a hair behind the best one is counted as good, at no cost. You see the full mix on every decision, so you can tell a fine play in a mixed spot from a real error.
What isn't graded
We label these instead of guessing:
- Multiway postflop pots.
- Heads-up tables, straddles, antes and tables with more than six players.
- Preflop lines that leave the chart, such as a raise far from any size the chart has, and every decision after that point.
- Chart spots the solve never settled on.
- Games other than No-Limit Hold'em cash.
Each one carries the reason on the decision, so you know why it has no grade.
Where the grades go
Grades roll up into the rest of the app. Results lists your hands costliest first. Leaks groups the decisions you repeat by spot, compares your frequency with the chart's for the cards you held, and ranks each leak by bb/100. The Trainer builds drills from your own mistakes. For a worked example on real data, see the preflop leaks in a real database of 14,124 hands.
Drop in a session and see every decision graded, with the cost in big blinds.
Grade your hands free