Poker Solver Glossary

Short definitions of the terms used in GTOkiller and in exploitative, GTO and population-data study. Each entry is self-contained.

Every definition below is written to stand on its own in 40 to 60 words, so it can be quoted directly. Terms marked with the GTOkiller label describe how this solver uses the concept.

Exploitative solver: A solver that computes the strategy maximizing EV against a specific, imperfect opponent model instead of the equilibrium against a perfect one. The opponent is fixed with measured data and only the hero's side of the tree is optimized. GTOkiller uses the real player pool of a site, stake and profile as that model.; See also: GTO (Game Theory Optimal), MDA (Mass Data Analysis), Solid exploitation
GTO (Game Theory Optimal): The equilibrium strategy of a poker game: a pair of strategies in which neither player can increase their EV by deviating. Computed by algorithms such as CFR, it is unexploitable but not EV-maximizing against opponents who make mistakes. In GTOkiller it appears as a reference line next to the exploit.; See also: CFR (Counterfactual Regret Minimization), Nash equilibrium, Exploitative solver
Nash equilibrium: A set of strategies, one per player, in which no player can gain by changing their own strategy while the others keep theirs. In two-player zero-sum poker it guarantees at least the game's value against any opponent, which is why GTO solvers target it. Against an opponent who deviates, a better strategy always exists.; See also: GTO (Game Theory Optimal), CFR (Counterfactual Regret Minimization)
CFR (Counterfactual Regret Minimization): The iterative algorithm behind modern GTO solvers. It plays the game against itself many times, adjusting each decision to reduce the regret of not having chosen better actions, until the strategies converge to an equilibrium. Introduced by Zinkevich, Johanson, Bowling and Piccione in 2007.; See also: GTO (Game Theory Optimal), Nash equilibrium
MDA (Mass Data Analysis): The statistical analysis of millions of real hands from a specific population to measure how it plays: action frequencies, bet sizes and ranges by spot. The result is a model of the pool rather than of a single player. GTOkiller uses MDA as the input of its solver, not only as a report.; See also: Pool, Population tendencies, Showdown bias
Pool: The set of players who make up a given game: a site, a stake and, optionally, a player profile such as regulars or recreational players. A pool has measurable average tendencies that differ from other pools. GTOkiller solves one strategy per pool, called a meta.; See also: Meta, Population tendencies
Meta: In GTOkiller, a specific pool defined by site, stake and opponent profile, with its own dataset and its own solved decision tree. Metas are published one at a time when the sample is large enough. The current live meta is the GG Poker NL10 to NL50 regular pool.; See also: Pool, Sample threshold
Population tendencies: The systematic, measurable ways in which a pool deviates from equilibrium in a spot, such as folding too often to a continuation bet or under-bluffing the river. Because they are averages over many players, they are stable enough to be exploited without reads on any individual.; See also: Leak, MDA (Mass Data Analysis)
Leak: A systematic, suboptimal tendency in how a player or a pool plays a spot, measured as a deviation from equilibrium frequencies. Overfolding to a c-bet, calling too wide preflop or overbluffing the river are leaks. Each leak has an EV value that an exploitative strategy can capture.; See also: Population tendencies, Exploit
Exploit: A deliberate deviation from the equilibrium strategy that gains EV against a known tendency of the opponent. In GTOkiller every exploit is visible: the three-strategy view shows the exploitative line, the GTO line and the population frequency it attacks, and the AI Coach names the EV gained.; See also: Leak, Three-strategy view, Solid exploitation
Solid exploitation: An exploitative strategy that attacks structural leaks while keeping a balanced base wherever the pool's deviation is marginal or the exploit would be fragile if the opponent adapted. It trades a small part of the maximum exploit for robustness against counter-adaptation. It is the output policy of GTOkiller's engine.; See also: Exploit, Exploitative solver
Node locking: Fixing by hand the strategy or frequency of one node in a GTO solver so that it solves assuming that behaviour. It exploits one decision point while the rest of the tree stays at equilibrium, and the solver typically assumes perfect play on later streets, so the resulting exploit is partial.; See also: Exploitative solver, Player profile
Player profile: In GTO solvers, a theoretical archetype of opponent, such as a calling station or a maniac, defined by hand-set incentives that bias the solver's assumed opponent. Profiles are not measured from real hands. GTOkiller replaces them with the measured frequencies of the actual pool.; See also: Node locking, MDA (Mass Data Analysis)
Three-strategy view: GTOkiller's presentation of every decision node: the exploitative recommendation, the GTO equilibrium line and the real frequencies of the population, side by side. It makes each exploit verifiable, because the user can see exactly which population tendency the recommendation attacks and by how much.; See also: Exploit, AI Coach
AI Coach: GTOkiller's on-demand explanation of a recommendation, in three steps: the spot and the highest-EV line, the EV breakdown that names the leak with its numbers, and the generalizable lesson. Every statement is anchored to a figure of that node; the coach does not reason freely.; See also: Three-strategy view, EV (Expected Value)
EV (Expected Value): The average amount a decision wins or loses over all possible outcomes, weighted by their probability, usually expressed in big blinds. A solver compares the EV of every available action in a node and recommends the highest. An exploit's value is the EV difference between the exploitative and the equilibrium line.; See also: bb/100, AI Coach
bb/100: Big blinds won per 100 hands, the standard measure of a poker player's win rate. It normalizes results across stakes. Small improvements matter: one extra bb/100 at NL50 is worth roughly 50 to 100 EUR per month for a regular playing 30,000 hands.; See also: EV (Expected Value)
Decision tree: The complete map of a hand from the flop onward: every action available to each player, every bet size, every turn and river card, and every resulting node. A GTOkiller meta contains about 500,000 nodes, each carrying the pool's real action frequency and sample size.; See also: Node, Sample threshold
Node: A single decision point in the decision tree, defined by the sequence of actions and cards that led to it, the pot and the player to act. In GTOkiller each node stores the real frequency of every action of the pool, the sample size and the solved exploitative strategy.; See also: Decision tree
Sample threshold: The minimum number of observed hands a node needs before its data and strategy are published. In GTOkiller it is dynamic: nodes whose strategy depends on range composition require more hands than nodes decided by frequencies alone. Nodes below their threshold are discarded, never filled with theory.; See also: Node, Showdown bias
Showdown bias: The selection effect that arises because only hands reaching showdown reveal their cards, so a range reconstructed from revealed hands overweights strong holdings. Action frequencies are not affected; range composition is, and GTOkiller corrects it with inverse probability weighting and Bayesian estimation.; See also: MDA (Mass Data Analysis), Sample threshold
Range: The set of hands a player can hold at a given point, each with a weight between 0 and 1 that reflects how often the player takes the line with that hand. Population ranges are the measured ranges of a pool by position and spot; GTO ranges are the equilibrium ones.; See also: Node, MDA (Mass Data Analysis)
C-bet (continuation bet): A bet on the flop by the player who made the last raise preflop, continuing the aggression. The frequency with which a pool c-bets, and the frequency with which it folds to a c-bet, are among the most exploitable population tendencies at low stakes.; See also: Leak, Population tendencies
3-bet: The second raise preflop: a re-raise of the initial open raise. A 3-bet pot is a hand in which the 3-bet was called. Pools 3-bet with different frequencies and ranges by position, and defend against 3-bets with measurable, exploitable tendencies.; See also: Range, Population tendencies
Overfold / overbluff: Folding more often, or bluffing more often, than the equilibrium frequency in a spot. An overfolding pool is exploited by bluffing more; an overbluffing pool is exploited by calling more. Both are measured directly as action frequencies, which are not affected by showdown bias.; See also: Leak, C-bet (continuation bet)

About GTOkiller: methodology and data · GTOkiller vs GTO Wizard