Read the assumptions first.
A result is useful when you can reconstruct the question it answers. Our articles separate configured numerical solutions, engine estimates, and illustrative arithmetic.
Describe the game before the result
Record the players’ positions, preflop action, starting and remaining effective stacks, pot, board, and both weighted ranges. State whether rake or antes are included. A label such as button versus big blind does not specify these inputs by itself.
Record the available actions on every relevant street, including permitted raises and all-ins. A simplified menu can help answer a bounded research question, but its output should not be presented as a recommendation over sizes it never considered. PioSOLVER’s tree-building documentation provides a concrete example of the settings a postflop model needs.
Separate the error measures
A numerical solver’s Solve tolerance concerns convergence inside the configured game. A learned search’s measured decision loss compares its strategy with a stated reference. Neither number measures whether your opponent follows the input range.
When publishing a benchmark, identify the model version, reference method, sample selection, sample size, units, and aggregation. Report weak spots alongside the mean, and distinguish the reference solver’s residual from approximation error. More training data or more search rounds are changes to test, not proof of improvement.
Interpret values and frequencies consistently
An expected-value comparison needs a common accounting convention. Values measured from the current decision cannot be mixed with values that include earlier investments. Record the pot or stake unit rather than leaving a number unlabelled. PioSOLVER’s explanation of its displayed values illustrates why conventions matter.
A Action frequency needs a denominator: a particular hand, a weighted range at a decision, a set of board patterns, or observed opportunities. Those are different averages. Our article on a big-blind check uses a deliberately small hypothetical range to show the conditioning step without inventing solver output.
Publication and review
Articles carry publication and review dates, source links, and limitations beside the claims they qualify. Tool guides identify documentation checks separately from firsthand tests. Changing prices and access conditions are checked at publication; an inaccessible price is left unquoted.
Turnstone and SolvedHand have the same owner. That relationship is disclosed beside recommendations involving SolvedHand and in the editorial policy. It is not evidence of comparative accuracy. Our tool-selection guide applies those standards to a bounded list of study tasks.