PulseGTO

Getting started

Reading strategies, frequencies and EV

The range grid

The 13×13 grid shows the player to act. Pairs are on the diagonal, suited hands above it and offsuit hands below it. Three views colour the grid:

  • Strategy: each cell shows the mix of actions for that hand, for example 60% bet and 40% check.
  • Equity: the hand's equity against the opponent's range at this node.
  • EV: the hand's expected value in big blinds.

Range shows one grid; Split shows both players side by side.

Frequencies: COMBO, REACH, FULL

  • COMBO: percentages per combo. "AKs bets 60%" means each AKs combo at this node bets 60% of the time.
  • REACH: weighted by how often each hand actually arrives at this node.
  • FULL: relative to the player's whole starting range.

The action totals above the grid use the same setting. Under REACH, for example, "bet 35%" means 35% of the range that reaches this node bets.

EV

EV is shown in big blinds for the player to act, counted from this decision onward. With Action EV switched on, each action button shows the EV of taking that action. The strategy detail shows the same numbers as a table, by rank, as a graph, or as an EV curve.

Blueprint postflop EVs are approximate

In a blueprint, postflop EVs are estimates from an abstraction-based training run, not values calculated separately for every exact combo on the current board. The app marks this:

  • In the combo table, blueprint postflop EVs are shown in italics, with a note that they are estimates from training, not per-combo values.
  • EQR is not computed from them.

How to read them. A blueprint EV is the value training recorded for a group of similar hands on boards of the same texture. It is useful for the broad picture: which lines are worth more, and roughly how much a range makes. It is not the exact value of one combo on this board, so it should not decide between two actions that are close. For that, re-solve the node: a RangeCFR re-solve gives per-combo EVs without hand abstraction. Their precision still depends on how far the solve has converged.

Across the tested flop spots, blueprint EVs differed from dedicated exact re-solves by about 0.05 bb per hand on average. Single combos and single spots can differ more, which is why these values are marked as estimates. The blueprint's frequencies are the more reliable part: they are what the training optimises.

  • Heads-up re-solves show per-combo EVs, computed with everyone playing the re-solved strategy. How close they are to equilibrium values depends on the re-solve's accuracy. They are checked combo by combo against the reference solver.
  • Three-way re-solves show per-combo EVs after the solve's EV evaluation completes. Their accuracy depends on convergence; a reduced-turn flop preview does not cover turns outside its sample.
  • Preflop action EVs are accumulated while the preflop engine trains the whole game, including the postflop play reached after an opening action. They estimate action values for each starting-hand class under the trained strategy. The engine aims at an equilibrium of the configured game; convergence and the chosen abstraction still affect the values.
Important

Which postflop EVs a blueprint has depends on the Postflop EV setting used in training:

  • in training: EVs are accumulated during training.
  • sampled at end: EVs come from a sampling pass after training, and can be missing where no sample reached a node.
  • off: postflop EVs are not recorded.

Blueprints trained before these modes existed got their postflop EVs from an older end-of-run pass, and those EV figures are less reliable. Their frequencies are not affected.

Runouts, hand classes and Scared Off

  • Runouts: how the strategy and EV change for every possible next card.
  • Hand classes: the range split into made hands and draws on the current board.
  • Scared Off: the opponent's range at this node, split into hand classes on the current board. If you pin a hand, it also shows how much of each class beats that hand at showdown right now, with the cards of your hand removed from the opponent's range. In a three-way pot the headline gives each opponent separately; the class list combines both. It describes the range as it is now; it does not predict future cards or folds.

Compare an action gap before comparing frequencies

For an arithmetic example, take a hand with Call worth 0.100 bb and Raise worth 0.102 bb under one fixed evaluation. A 50/50 mix gives 0.101 bb; always calling gives up 0.002 bb against the best action. Large frequency differences can therefore have small value costs. This example is not an output from a particular hand or a new benchmark.

Use the same EV baseline: a value relative to folding can differ from a net value including a posted blind by a constant. Also separate values recorded while policies were changing from a fresh evaluation of the finished policy. Reading an old blueprint EV beside its averaged strategy does not make the EV a final-policy evaluation.

At a locked hand, first check which actions are allowed. An excluded action is not a valid recommendation just because its unrestricted EV is higher. In a postflop spot, re-solve per combo and inspect convergence before treating a tiny action gap as decisive.