Expected value, or EV, is the average amount an action wins or loses across repeated outcomes under a defined model. It is the clearest way to compare poker actions when the immediate result is affected by chance.
EV is an average, not a prediction
An action with positive EV can lose in one hand. A negative-EV action can win. EV describes the long-run value of the decision, not what must happen next.
Relative EV in training feedback
A solver may display the value of each available action. The highest-EV action is the reference. The gap between your action and the best action measures the cost of the mistake under the model.
Why frequency mistakes can have small EV loss
When two actions mix, their EVs are often close. Selecting one too frequently can be strategically imperfect without being expensive on that exact hand. This is why EV-based review is more informative than an error count alone.
Large EV gaps identify urgent leaks
If raising and folding have very different values, choosing the wrong action is more serious. Repeated large gaps in one position or category should receive priority in study.
bb/100 as a training display
Normalising EV to big blinds per 100 hands can make small per-decision differences easier to read. It remains a model output for the specific decision, not a direct prediction of your overall win rate.
| Term | Meaning |
|---|---|
| Best-action EV | Value of the highest-EV available action |
| Chosen-action EV | Value of the action you selected |
| EV loss | Difference between chosen and best action |
| Cumulative EV loss | Sum of modelled losses across trained decisions |
| Accuracy | Share of answers classified as correct under the training rule |
Use EV without ignoring assumptions
The value depends on ranges, stack sizes, rake, bet sizes and actions included in the model. Compare decisions within the same solved configuration and avoid treating one number as universal.
Frequently asked questions
What does positive EV mean?
It means the action earns value on average under the assumptions and repeated outcomes represented by the model.
Can a correct action lose money in one hand?
Yes. Variance determines individual outcomes. EV evaluates the decision across many possible outcomes.
Why can an incorrect training answer lose almost no EV?
The hand may sit in a close mixed-frequency region where two actions have nearly equal value.
Is training EV the same as my actual win rate?
No. It measures decisions in the configured training spots. Your win rate also includes postflop play, opponents, rake, game selection and variance.