ConceptsFoundations

EV, EV loss and how Rail grades

Expected value is a decision's average result; Rail grades a decision by how much EV it gave up against a reference strategy.

What it is

The expected value (EV) of an action is what it wins or loses on average, over every way the hand can continue, in big blinds. Folding is worth 0 from the moment you fold; every other action is measured against that.

EV loss is the gap between the best option and the one you chose: max EV − EV of your choice. Zero means you chose an option as good as the best; it is never negative.

Rail grades a decision by comparing it with a reference strategy — preflop charts, and later a solver. The reference gives every option at the spot a frequency (how often it plays that option with your exact hand) and an EV. Where there is no reference, Rail does not grade: it shows flags, notes about checks that hold whatever the strategy, and never calls a flag worse than Inaccurate.

Why it matters

Results lie in the short run. Winning a pot with a bad call does not make the call good; EV is how you tell the decision from the luck.

Rail measures EV loss as a share of the pot, not in big blinds, because the same 2 bb is a disaster in a 4 bb pot and noise in a 100 bb pot.

Strategies mix. When the reference calls 52% and raises 48%, both are right, so Rail treats every option within 5 percentage points of the most frequent one as Perfect. A player who always picks the 48% option is not wrong in any single hand; that pattern shows up in reports, never as a grade.

Each move also gets a score from 0 to 100, for averages: 100 minus 1,000 times the EV loss as a share of the pot, so losing 3% of the pot scores 70 and losing 10% or more scores 0. When the loss is negligible, the score is the option's frequency against the most frequent one's.

The grades, in order — Perfect: within 5 points of the most frequent option, or losing at most 0.1% of the pot. Good: the reference plays it at least 3.5% of the time. Inaccurate: rarer than that, losing at most 2% of the pot. Mistake: losing at most 8%. Blunder: anything more.

The maths

EV loss equals the highest EV among the options minus the EV of the option you chose.

EV loss as a share of the pot equals the EV loss divided by the pot before the decision.

The move score equals 100 minus 1000 times the EV loss as a share of the pot, and never less than zero.

Worked example

Grading one flop decision

The pot is 10 bb and you face a bet. The reference with your hand: fold 0% (EV 0), call 52% (EV 1.20 bb), raise 48% (EV 1.17 bb).

You raise. 48% is within 5 points of 52%, so it is Perfect. The EV loss is 0.03 bb, 0.3% of the pot, so the move scores 100 − 1000 × 0.003 = 97.
You fold instead. The reference never folds, and the EV loss is 1.20 bb, 12% of the pot: a Blunder, scoring 0.
Change the spot: the reference raises only 2% of the time, with EV 1.05 bb. Raising now loses 0.15 bb, 1.5% of the pot, and is rarer than 3.5%: Inaccurate, scoring 85.

The grade follows how much a choice costs and how often the reference makes it — never whether the hand was won.

Try it

Set the reference frequencies and EVs, pick the option you played, and watch the grade and score change.

Pot before the decision10 bb

Reference frequency of call52%

Reference frequency of raise48%

Fold takes the rest: 0%

EV of call in bb+1.2 bb

EV of raise in bb+1.17 bb

You playedFoldCallRaise

Grade: Perfect
EV loss: 0.03 bb · 0.3% of the pot
Score: 97

Common mistakes

Judging a decision by the result of the hand.
Reading "Good" as "wrong". Good is a choice the reference makes too, just not the most often.
Comparing big-blind losses across pots of different sizes instead of the share of the pot.
Treating a heuristic flag as a grade. A flag says a check failed; only a reference strategy can say how much it cost.

In your hands

Hands from your library where the analysis flagged this:

Opens your analysis, filtered. You need to be signed in; nobody else sees your hands.

All concepts