// variance · stated model

How many hands does it take to know your win rate?

The width of a measured win rate by sample size, and the hands a given true win rate needs before the result is ahead with confidence.

Over n hands your measured win rate is the true one plus noise, and the noise has a standard deviation of s / √t, where t = n / 100. A 95% interval for the measured figure is the true win rate ± 1.96 s / √t.

s = 90 bb/100, n = 50,000 hands, t = 500
± 1.96 × 90 / √500 = ±7.9 bb/100

A measured +3 over 50,000 hands is consistent with a true win rate
anywhere from −4.9 to +10.9 bb/100 at 95%.

How wide the measurement is

Standard deviation 90 bb/100. The ± applies to whatever the true win rate is.

Half-width of the 95% interval on a measured win rate by hands played
Hands played95% interval (± bb/100)
10,000±17.6
25,000±11.2
50,000±7.9
100,000±5.6
250,000±3.5
1,000,000±1.8

Hands until a win rate shows up

The other direction: how many hands before a player whose true win rate is w finishes ahead with chosen confidence. The formula is 100 × (z s / w)², with z the normal quantile for the confidence level (1.282, 1.645, 2.326).

Hands needed to be ahead with 90, 95 and 99 percent confidence by win rate
Confidence of being ahead1 bb/1002 bb/1003 bb/1005 bb/1008 bb/100
90%1,331,000333,000148,00054,00021,000
95%2,192,000548,000244,00088,00035,000
99%4,384,0001,096,000488,000176,00069,000

Rounded up to the nearest thousand. Being ahead is a low bar: it says the result is above zero, not that the win rate is the one you hoped for. Pinning a win rate down to ±1 bb/100 at 95% takes about 3,100,000 hands at this standard deviation. This is also a statement about a single player’s sample, not about whether any given player has an edge: the page cannot tell you that, and a small sample cannot either. Try your own numbers.

More on variance

Variance calculatorRisk of ruinBankroll in buy-insDownswingsPot odds

// what is computed and what is assumed
  • The arithmetic is exact; the inputs are yours. Every figure follows from the win rate and standard deviation you supply, by the formula printed beside it. This site cannot tell you your win rate, and a positive input is not a promise that you have one.
  • The win rate is treated as known. In practice it is the biggest unknown. The how many hands page shows how wide the range of a measured win rate is, even after a large sample.
  • Normal model. Results are modelled as a random walk with a fixed mean and standard deviation per 100 hands. Real graphs are lumpier than that: a hand’s result is not normally distributed, and a player’s win rate changes with stakes, game selection and form. The closed forms were checked against a hand-by-hand simulation of the same model, not against anyone’s tracked results.
  • The default standard deviation, 90 bb/100, is an assumption, not a measurement. Figures in the region of 80 to 100 are often quoted for 6-max no-limit hold’em cash games; we have not measured one. Use the standard deviation from your own tracking software, and the calculator will use it instead.
  • Cash-style results only. A buy-in here is 100 big blinds unless you change it. Tournament results are far more skewed than this model allows, so nothing here should be used to size a tournament bankroll.
  • Nothing is sent anywhere. The calculator runs in your browser; the URL carries your inputs so you can share a scenario.