Game Tree

Detailed Explanation

A game tree is the complete mathematical representation of every possible action sequence, decision point, and outcome in a poker hand or game. In its most fundamental form, a game tree begins at an initial state (such as cards being dealt) and branches outward with each possible action available to players, creating nodes that represent game states and edges that represent player actions. This structure continues recursively until reaching terminal nodes where the hand concludes and payoffs are determined.

In poker solver technology, the game tree serves as the foundational framework upon which equilibrium strategies are computed. Modern GTO solvers like PioSolver, GTO+, and Simple Postflop construct these trees by defining parameters including stack depths, bet sizes, raise sizes, and which parts of the game tree to solve. The complexity of a poker game tree grows exponentially with each decision point—a single postflop spot in No-Limit Hold'em can contain billions or trillions of nodes when accounting for all possible card runouts and action sequences.

Each node in a game tree contains critical information: the current game state (including pot size, stack sizes, board texture, and action history), whose turn it is to act, what actions are available, and the range distributions of all players at that point. Chance nodes represent random events like card dealing, while decision nodes represent points where players choose between actions. The solver's objective is to assign probabilities to each available action at every decision node such that no player can exploit another's strategy—the Nash equilibrium.

The construction of an accurate game tree requires careful abstraction, as solving the complete game tree for even a single poker hand would be computationally infeasible. Practitioners must make strategic choices about which bet sizes to include (discrete betting options versus continuous sizing), how finely to bucket similar hands together, and which branches merit deeper investigation. These abstractions create a simplified game tree that approximates the full game while remaining computationally tractable.

Practical Examples

Consider a simplified single raised pot (SRP) scenario where the button opens and the big blind calls. On an A♠K♣7♥ flop, the game tree branches from the initial node where both players have checked their options. At this decision node, the big blind can check or bet. If we allow three bet sizes (33%, 75%, and 150% pot), the tree now has four branches from this node.

If the big blind checks and the button bets 75% pot, we reach a new decision node where the big blind must respond. Available actions might include fold, call, raise to 3x, and raise all-in. Each of these creates a new branch. If the big blind raises to 3x and the button calls, we proceed to a chance node where the turn card is dealt. With 47 unknown cards remaining, this creates 47 new branches, each leading to another decision node where the big blind acts first.

The tree's complexity multiplies dramatically with each street. A three-bet pot on a K♠Q♦6♣ flop with 150bb stacks might allow for continuation bets of 33%, 50%, and 75% pot, then facing a raise to 2.5x or 4x, then facing a 3-bet shove or call, proceeding through turn and river with multiple sizing options at each stage. A complete game tree for this scenario, accounting for all possible turn and river cards and action sequences, could easily contain hundreds of millions of nodes.

In a tournament setting with 25bb effective stacks, the game tree simplifies considerably. Preflop from the small blind versus button, if we restrict actions to fold, call, or shove, each decision point has fewer branches. The shallower stack depth means postflop trees also have fewer viable bet sizes, and many hands will be all-in by the turn or river, creating terminal nodes earlier and reducing overall tree complexity.

Strategic Considerations

Understanding game tree construction directly impacts how effectively you can use solver output in your strategy development. When building custom solutions, the bet sizes you include in your game tree should reflect the actual decisions you face at the table. Including too many sizing options creates computational burden without strategic benefit, while including too few may miss critical parts of optimal strategy. Many professionals solve with 2-4 bet sizes per street, focusing on geometrically spaced options that serve distinct strategic purposes.

The concept of subgame solving leverages game tree structure to enable real-time adjustments. Rather than memorizing an entire precomputed game tree, advanced players identify the current node they've reached during a hand and resolve the subtree from that point forward using updated range information. This approach allows for exploitation of opponent tendencies while maintaining the structural framework of GTO play.

Pruning strategies represent another critical consideration. In game tree construction, you can exclude branches that are unlikely to occur in practice. For example, if you determine that betting 200% pot on the flop serves no strategic purpose for either range, you can remove this branch entirely, dramatically reducing tree size. However, aggressive pruning risks eliminating important strategic options, so skilled practitioners balance computational efficiency with strategic completeness.

The game tree framework also illuminates why certain poker concepts exist. The value of position becomes apparent when examining how the tree's structure changes based on who acts first—the in-position player reaches many more favorable decision nodes where they act with information about their opponent's action. Similarly, concepts like range advantage and nut advantage can be quantified by analyzing which player's range contains stronger hands at specific nodes throughout the tree.

Common Misconceptions

A prevalent misconception is that solvers compute optimal strategy by evaluating every possible hand individually at each decision point. In reality, solvers operate on the entire game tree simultaneously, using algorithms like counterfactual regret minimization (CFR) to iteratively improve strategies across all nodes until convergence. Individual hand decisions emerge from this holistic optimization rather than being calculated in isolation.

Many players incorrectly assume that the game tree used by a solver represents absolute poker truth. However, the game tree is an abstraction based on the builder's choices about which actions to include. A solution from a tree that includes only 33% and 75% pot bets may differ significantly from one that includes 50% and 100% pot bets. Neither is "correct" in an absolute sense—they answer different questions based on their construction parameters.

Another error involves conflating the game tree with a decision tree. While related, a game tree includes chance nodes for random events (card dealing) and models the strategic interaction between multiple players, whereas a decision tree typically represents one player's choices. This distinction matters because optimal play at any node depends on how opponents will respond at future nodes throughout the tree.

Some practitioners mistakenly believe that memorizing solver outputs from a precomputed game tree will allow perfect play. However, live play inevitably deviates from solved scenarios due to different stack depths, opponent actions not included in the tree, or card runouts that reach nodes you haven't studied. Understanding the principles behind why the solver makes certain choices at various nodes proves more valuable than rote memorization of specific frequencies.

Finally, players often underestimate how dramatically game tree complexity increases with stack depth. A 40bb game tree might solve in minutes, while a 200bb tree with similar parameters could take days or prove unsolvable on consumer hardware. This exponential growth in complexity explains why professional solvers focus heavily on optimization and why deepstack play remains less solved than short-stack scenarios.