Analytics and decision review
11.1 What is a decision point?
A decision point is any state where you choose among legal actions: check or bet; call, raise, or fold; select a size; or choose an all-in. Not every decision point is equally informative. The most valuable analysis targets are spots where the EV ranking of alternatives is unclear, where a repeated leak may exist, or where large pots magnify small frequency errors.
11.2 The review loop
- Collect hand histories or write down live hands while memory is fresh.
- Filter for one family of decisions: for example, big-blind calls versus button opens, turn continuation bets, or river bluff catches.
- Reconstruct the information state: positions, stack depth, sizes, board, and plausible ranges.
- Create a mental model before looking at the result. Estimate equity, thresholds, and likely opponent responses.
- Use software, enumeration, or a solver when the model is too complex for hand calculation.
- Compare your policy to the analysis, classify the error, and write one transferable rule or question for future play.
- Repeat on a sample, not a single dramatic hand.
The anti-results rule
Do not begin review with "I lost a big pot with aces." Begin with a decision class: "I faced a turn raise after betting an overpair on coordinated boards." Outcomes are noisy; decision classes are analyzable.
11.3 Sample size
Many poker statistics converge slowly. A player can run far above or below expectation over thousands of hands in all-in outcomes, and rare actions such as four-bets may have tiny samples. Treat observed frequencies as estimators with uncertainty. A 70% fold-to-three-bet rate over 10 opportunities is weak evidence; the same rate over 500 opportunities is qualitatively different.
11.4 Useful metrics, carefully
| Metric | What it measures | Common misuse |
|---|---|---|
| VPIP | Fraction of hands where a player voluntarily puts money in preflop. | Treating it as hand quality rather than looseness. |
| PFR | Fraction of hands raised preflop. | Ignoring position-specific differences. |
| 3-bet | Frequency of re-raising preflop. | Using tiny samples or mixing all positions. |
| C-bet | Postflop bet by the preflop aggressor. | Ignoring board class, size, and number of opponents. |
| WTSD | How often a player reaches showdown after seeing a flop. | Assuming a high number means one specific leak. |
Metrics are features, not diagnoses. Strong analysis conditions on the state. The more heterogeneous the pooled situations, the less a single percentage means.
11.5 Separate model error from execution error
A useful postmortem classifies mistakes. Did you misunderstand the rule or formula? Misestimate the opponent range? Choose the wrong size? Fail to execute a known policy under time pressure? Or simply encounter a low-probability outcome? Each category implies a different remedy.