This article is a methodology lesson, not a product ranking. It separates a solver from an equity calculator or hand-history tracker, traces a small hand-calculable toy game, and lists what a comparison would need before claiming one tool is better. In-play use is a room-policy question; the fact that software can solve a spot does not mean an operator permits consulting it during play.
Solver, Equity Calculator, Tracker and Decision Aid Are Different Tools
An equity calculator estimates a hand's expected share of the pot against specified hands or ranges over possible runouts, including split pots, if the hand reaches showdown without further betting. A tracker organizes past hand histories and statistics. A solver includes possible decisions and responses, then seeks a strategy within that game tree. A real-time decision aid is a usage pattern or feature that informs a current hand; it may use solver output, a chart or other analysis, and can be restricted by a room. These categories can overlap in products, but they are not synonyms.
| Question | Typical tool class | What the answer does not establish |
|---|---|---|
| “What share of the pot would this hand claim against that range at showdown?” | Equity calculator | Which bet size or action maximizes EV. |
| “How did I play similar hands last month?” | Tracker or database | That the recorded sample predicts the next opponent. |
| “How should both players mix actions in this modeled tree?” | Solver | That the real opponent uses the assumed ranges or sizes. |
| “What should I do on this live hand?” | Real-time advice workflow | That the operator allows the workflow. |
Poker-Eye's product pages should not be described as a solver or equivalent to a named solver without a separately tested feature contract. A strategy article does not grant platform compatibility or operator approval.
The Inputs Define the Question Being Solved
For a postflop no-limit Hold'em analysis, a solver needs starting ranges for both players, the board cards, pot and effective stacks, permitted bet and raise sizes, and a utility treatment such as cash-game rake or tournament payout model when applicable. Their relative positions determine who acts first at each decision. GTO Wizard's explanation lists these as game-space inputs and describes how limiting bet sizes makes a huge poker game computationally manageable. If you leave out a size that a real player can choose, the calculated strategy cannot literally cover that omitted branch.
| Input | Example value in a hypothetical study | If it is wrong or omitted |
|---|---|---|
| Player ranges | BTN open versus BB call | The solution evaluates the wrong possible hands. |
| Board | K♣7♦2♠ | Hand strengths and blockers differ on another flop. |
| Pot and effective stack | 20 BB pot, 80 BB behind | Bet sizes and future commitment change at another depth. |
| Action tree | Check, 33% pot bet, 75% pot bet, allowed raises | A full-pot action has no branch if it was excluded. |
| Utility | No-rake chip EV, or a specified rake/payout model | Cash rake or tournament payouts can change the objective. |
These example inputs are not a computed solution. They are a checklist of what a real solved spot would have to disclose. The opening ranges themselves need provenance and a version. A reader should be able to say exactly which tree produced the color on a strategy grid.
A Decision Tree Is a Map of Available Actions
A node is a decision point; a branch is an action; a terminal node ends in a fold or showdown with a defined payoff. A two-player river tree might allow one player to check or bet, then the other to call or fold. Add a second bet size, a raise option or another street and the number of branches grows. A solver assigns strategies at information sets—situations where a player has the same information—and evaluates how the players' choices interact.
At each terminal branch, expected value is the payoff multiplied by the probability of reaching it. Sum over branches for an action. GTO Wizard's article explains this terminal-payoff calculation and discusses iterative regret reduction. The original 2007 counterfactual-regret paper is a primary research source for one important family of methods, not a claim that every commercial solver has identical internals.
A Small River Game You Can Calculate by Hand
To see why a strategy can mix actions, use a toy game, not a claim about a real Hold'em board. There is 100 chips in the pot. An attacker can hold a value hand that always wins if called or a bluff hand that always loses if called. The attacker may bet 50 chips or check. A defender has a bluff-catcher that beats every bluff and loses to every value hand; facing a bet, the defender may call 50 or fold. There are no ties, raises, rake or future streets. Assume the value hand is always bet and the bluff hand would win zero additional chips by checking. The value and bluff hand types are initially equally frequent in this deliberately constrained model.
When a bluff bets, the defender folds with probability 1 − c or calls with probability c. The bluff wins the existing 100 if the defender folds and loses its 50 bet if called. Its EV relative to checking for zero is:
bluff EV = (1 − c) × 100 + c × (−50) = 100 − 150c
The bluff is indifferent between betting and checking when c = 100/150 = 2/3: the defender calls 66⅔% of the time. That is a result of this fixed 100/50 tree, not a general “always call two-thirds” instruction.
The defender's call costs 50. If the bet is a bluff with conditional probability q, calling wins the 150 already in the pot after the attacker bets; if it is value, calling loses 50. Relative to folding:
call EV = q × 150 + (1 − q) × (−50) = 200q − 50
The defender is indifferent when q = 50/200 = 25%. If the attacker's value and bluff types are initially equally frequent and all value hands bet, betting the bluff type only one-third of the time creates betting mass 0.5 value + (0.5 × 1/3) bluff = 2/3, of which bluff mass 1/6 is one quarter. In this constrained toy tree, a one-third bluff frequency and two-thirds defender call frequency satisfy both indifference equations. This is not a solved multi-street poker strategy; it simply illustrates why a model may output mixed percentages rather than one pure button press.
| Toy input or output | Value | Recalculate |
|---|---|---|
| Pot before bet / bet / call | 100 / 50 / 50 | Final pot after call = 200. |
| Defender's break-even bluff share among bets | 25% | 50 / 200. |
| Defender call frequency making bluffs indifferent | 66⅔% | 100 / (100 + 50). |
| Attacker bluff-type bet frequency under 50/50 initial types | 33⅓% | (0.5 × 1/3) / (0.5 + 0.5/3) = 25%. |
What “Mixed Strategy” Really Means
A solver may assign a hand a 70% bet and 30% check at one node. That does not mean it knows the next opponent's exact response. It means that, within the solved model, mixing those actions can make the player's strategy hard to improve against or reflect near-equal values for alternatives. Frequencies can be sensitive to tiny EV differences, tree abstraction and convergence tolerance. A novice who reads “70% bet” as “bet every time” has changed the strategy; a writer who calls it “optimal in all poker games” has changed the claim.
The toy tree above is intentionally transparent. Real Hold'em has many hole-card combinations, possible future boards, bet sizes and opponent responses. A commercial solver may use different algorithms or abstractions. The GTO Wizard methodology discusses how a tree that is too narrow can let the solution exploit its own artificial limits, while a tree that is too large becomes costly to compute. An output screenshot without the tree specification is not a reproducible study.
Abstraction, Convergence, Rake and Real-Table Uncertainty
Abstraction reduces the game to a tractable set of cards or actions. Convergence describes how close an iterative solution is to its chosen stopping criterion. The output may be precise to one decimal place but still belong to an oversimplified tree. Rake, effective stacks, payout rules, ranges and bet sizes can all change the EV landscape. Treat a difference of a tiny fraction of a chip between two actions cautiously unless the source exposes its accuracy and assumptions.
Guerrilla Poker sketches how a solver iteratively adjusts responses in English. English captions are available. This is for off-table study; live use depends on each room's rules.
Even an excellent solution to its own game does not measure how frequently a particular opponent bluffs, whether the room's rake matches the model, or whether a player can execute a mixed action. Off-table study can use the result to ask why an action appears; it should not be presented as an automatic real-table command. Our GTO poker guide covers the strategic term; this article owns the input-output audit.
Preflop and Postflop Are Different Computational Tasks
Preflop modeling must account for a much larger tree of future boards and decisions. Postflop analysis starts from specified ranges and a known flop or later board, reducing the state space but making the answer sensitive to how those input ranges were constructed. A preflop opening chart is not an answer to a river bet on a particular board. Conversely, a postflop solve that begins with the wrong preflop ranges can be internally consistent and still answer the wrong question.
When reviewing a vendor's output, check whether the solve is precomputed or generated for the current setup, what actions are available, how strategy frequencies and EVs are displayed, and whether the tool documents accuracy. We have not run a head-to-head vendor benchmark for this article. A feature list is not evidence of which vendor solves faster, more accurately or more usefully in an independently held task.
A Fair Comparison Protocol, Not a Fabricated Ranking
To compare solver classes or products, first specify one identical test task: variant, public board, both ranges, pot, stacks, bet tree, rake or payout model and target accuracy. Record tool name, version/build, machine, settings, run duration, memory use, exported strategy, EV and any warning. If a product cannot express the same tree, report not comparable for this task rather than forcing a winner. Have an independent reviewer reproduce inputs and outputs. Until that test is performed, the status of each product's numerical performance is not tested.
| Tool class | Test question | Source/version to record | Boundary to report |
|---|---|---|---|
| Precomputed solution library | Does it contain the exact specified scenario? | Library/version and tree metadata | A nearby spot is not the same solve. |
| Configurable postflop solver | Can it solve the supplied ranges, board and action sizes? | Build, settings, hardware and accuracy | Different trees invalidate direct speed/EV comparisons. |
| Equity calculator | What is showdown equity for specified hands or ranges? | Algorithm/build and cards/ranges | It does not optimize betting strategy by itself. |
| Hand-history tracker | What happened in recorded past hands? | Version, import and filter settings | Observed frequencies are not a solved equilibrium. |
Study Use Versus Live Use
A solver is not automatically permitted during online play. PokerStars' tool policy restricts advanced GTO solutions while its software runs. GGPoker prohibits external real-time assistance and lists solvers in its auxiliary-tool section. CoinPoker's dedicated tool policy restricts RTA, listed solvers and decision-assisting training tools while its client is open; its 2026 terms, page 21, also prohibit RTA and solvers during live play. These sections were rechecked on 22 October 2026; they do not approve any unnamed product.
The safe editorial message is simple: study context and live decision context are separate. Do not use an educational diagram, product label or vendor blog as permission to consult a solution on a running client. For a named feature, seek the operator's written answer and preserve it with version/date.
The hand-review guide provides an off-table question-and-record process before any solver model is built; it grants no permission to use a solver during play.
Quick Answers
Does a solver know my opponent's cards? No. It operates on possible ranges and the known board within its defined model; it does not reveal hidden cards.
Why do solver outputs show percentages instead of one move? Mixed actions can arise within the model. Our toy 100/50 river tree illustrates indifference; actual outputs also depend on ranges, action sizes and convergence.
Is an equity calculator a solver? Not by itself. Raw equity estimates a showdown pot share under specified hands or ranges; it does not account for future betting or how much of that share a player will realize. A betting solver also models choices and responses in a game tree.
Can I compare two solvers from screenshots? Not fairly unless both use the same task, tree, ranges, rake/payout, version and accuracy standard. Otherwise the screenshots may answer different questions.
Sources and Scope
- GTO Wizard: How Solvers Work — Explains game-tree inputs, terminal payoffs, iterative solving and abstraction limits. It does not validate our toy tree as a real-table solution.
- Zinkevich et al.: Regret Minimization in Games With Incomplete Information (2007) — Original counterfactual-regret research for one solver-method family; not evidence that every commercial solver uses identical internals.
- PokerStars: Prohibited Programs, Tools and Services — Defines this operator's restriction on advanced GTO tools while its software runs.
- GGPoker: Security & Ecology Policy — Defines its external-assistance and auxiliary-tool restrictions for live play; educational use is not blanket permission.
- CoinPoker: Third-Party Tools & Software Policies — The RTA and Training Tools During Play sections restrict solvers and decision-assisting study tools while the client is open. Rechecked 22 October 2026.
- CoinPoker: Terms and Conditions (2026), pp. 4–5, 21 — Pages 4–5 define gameplay-connected HUD restrictions and the other-platform HUD clause; page 21 addresses live RTA/solvers and HUD use. These are distinct from the policy FAQ's client-open wording. Directly rechecked 22 October 2026.
The 100/50 river toy game is an original constrained model with printed indifference equations. No commercial solver benchmark or multi-street strategy was measured.
About the Editorial Team
The Poker-Eye Editorial Team researches, checks, and maintains product and educational content. Sources and product boundaries are reviewed before publication.