PulseGTO
Solver engines
RangeCFR (three-way)
Three-way RangeCFR solves postflop spots for three players from the flop, turn or river. Enter your own three ranges, board, pot, stacks and bet sizes in Postflop Settings, or use Recalculate Node on a three-player blueprint node. It is the three-player counterpart to heads-up RangeCFR: each hand combination has its own strategy, and the solver works through the selected betting tree rather than relying on the blueprint's postflop hand abstraction.
Full-card solve by default
| Starting street | Future cards in the default solve | Hands |
|---|---|---|
| River | Board complete | Every combo in all three ranges |
| Turn | Every possible river | Every combo in all three ranges |
| Flop | Every possible turn and river when the tree fits in RAM | Every combo in all three ranges |
The engine accounts for blockers between all three hands and the board, and for side pots, folds, ties and rake. It reports strategy, EV and convergence progress for the configured spot. You can continue the run toward your accuracy target; a short benchmark is not an iteration limit.
On a flop, all turns is the default. The 12-turn setting is an optional quick preview. If a full tree exceeds available RAM, the engine can cover as many turns as fit and shows exactly which ones it solved. On an unsolved turn, start a fresh turn solve to cover every river. Four or more players are not supported by this postflop engine.
Technical notes
The Acc metric reflects convergence within the configured betting tree. If memory forces reduced turn coverage or you select quick preview, Acc covers the solved cards; it does not measure the effect of omitted turns. See Exploitability metrics and Sampling accuracy.
Overview page: Heads-up and three-way postflop solver.
A useful first three-way trial
Begin with a complete river board and three realistic ranges. Keep the betting menu modest, inspect the memory estimate and solve until the accuracy is suitable for the question. Check each player’s range rather than assuming the middle player behaves like the out-of-position seat of a heads-up pot.
Next, start from a turn with the same action model and compare the results by river. Only then move to a full flop tree with wider ranges and more bet sizes. This progression lets you distinguish a wrong input state from a memory limitation.
Label a quick preview explicitly. If only some turns were solved, there is no result for the omitted turns just because the root accuracy looks small. A heads-up node lock is also not an available three-way feature. Use the dedicated controls and preserve the actual coverage with any exported example.