Poker win rate, tracking and sample size
A poker win rate only becomes useful when the sessions behind it are complete, the sample is large enough and the review loop changes what you do next. This guide joins the measurement, tracking and study pieces into one system.
A bigger sample changes the interval.
Keep the result at +5 big blinds per hour. Change only the recorded hours and watch the uncertainty narrow.
The interval spans zero. This sample does not distinguish a positive mean from zero under this model.
95% interval = observed rate ± 1.96 × assumed one-hour standard deviation ÷ √hours
Four times the comparable recorded hours halves this model's margin. This is a normal approximation with independent, comparable hourly outcomes and stable variability. Changes in stakes, game quality, costs or your decisions can break those assumptions. When variability is estimated from a small sample, a different method may be needed.
A 95% confidence procedure covers the true mean in 95% of repeated samples under its assumptions. It is not a 95% probability about this particular fixed interval, a forecast of the next session, or proof that your edge will last. Never play extra hours just to reach a statistical threshold.
Include losses, retain duration and stake, and label estimated entries. A carefully selected winning sample is not repaired by a confidence formula.
Sources and method
- NIST: confidence interval with known standard deviationSupports the normal-model formula; the poker inputs here are hypothetical.
- NIST: interpreting confidence limitsExplains repeated-sampling coverage and uncertainty about a mean.
Original Overshove writing, arithmetic and diagram. Illustrative figures are not player records.
A short observation is a question, not a player label.
Keep what you saw separate from the story you tell about it.
A 37.5% observed frequency does not establish a fixed 37.5% tendency. The Wilson interval is wide, and its model assumes independent opportunities with the same underlying probability. Different positions, sizes, opponents and board types can undermine that assumption.
Keep a neutral observation such as “folded three of eight recorded opportunities; other conditions uncertain”. Alternative explanations include different hand strength, a changing table or missed opportunities in your notes. Avoid personal or identifying details, and follow the room or platform’s rules.
Check yourself: did three folds out of eight prove an exploit?
No. It provides a small observation under a chosen definition. It neither identifies the unobserved hands nor proves that a proposed betting adjustment has positive expected value. More play is not required to finish a sample.
For your own records, compare like stakes, cost definitions and time measures. A £100 result after extra costs over four playing hours is £25/hour; over six hours including travel it is £16.67/hour. Neither describes what the next session will earn.
Sources and method
- NIST: confidence intervals for a proportionSupports the Wilson method, calculated here with three successes, eight trials and z = 1.96. The poker observation and interpretations are original.
Fresh Overshove writing and examples. Numbers are illustrative, independently calculated and not player records. No quoted book text, copied hand, range chart or artwork.
Tom’s five-minute session log
Thomas Howcroft, Overshove’s founder, says he usually records a cash session within five minutes of finishing. The reason is practical: the start and finish times, money put in, cash-out and food or travel costs are still fresh. Five minutes is his routine, not a scientifically established deadline.
He began keeping a complete record after a run of losses made him question what playing was costing. Recording both winning and losing sessions gave him a clearer history than trying to remember the memorable nights. He now describes several years of tracking. A complete record is useful whether poker is a hobby or a more serious commitment; it does not establish a future winning rate.
Table profit, expenses and two different hourly rates
Table profit is cash-out minus all buy-ins. Rake taken from pots and tips paid from the playing stack already reduce the cash-out: subtracting them again understates the result. Only deduct additional costs that are not already included. Label your figures so a table-only result is not confused with a result after food, travel and other recorded costs.
Tom records food and travel, but says he does not currently record dealer or staff tips separately. That is a limit of his personal record, not evidence that those costs are zero. He says several years of expense tracking changed his understanding of what remained after playing costs. This is a personal observation, not a claim about every player’s expenses.
Save enough context to review a decision
Tom reviews hands with a poker coach, sometimes through a short discussion after play. He wants the positions, effective stacks, cards, action and bet sizes, plus observations about the opponent. Those details let him discuss a decision rather than merely report whether the pot was won. Describe observed actions; a label such as “professional” or “recreational” is not a substitute for a specific read.
If an exact size is forgotten, write an estimate and its unit: for example, about half the pot or approximately 12 big blinds. Distinguish your estimate from a recorded amount. Reconstruct the action after the session, without letting phone use distract from the game or break the venue’s rules. Do not record other players’ private identifying details.
The current Overshove workflow is session logging and results review. A dedicated Study area and player-read workflow are in development; this checklist does not mean those planned modules are available. Keep detailed hand records in your existing study method for now. A results graph cannot establish whether a particular decision was good.
For a step-by-step method, use the ten-minute hand review: rebuild one decision, price it and test the read that drives it.
A month of live poker is a rounding error
Online players settle win-rate arguments with hand counts. Live players cannot, because the hands arrive too slowly. At 25 to 30 hands per hour, a busy 60-hour month deals you 1,500 to 1,800 hands. A full year at 15 hours a week, call it 750 hours, still lands under 23,000 hands. In per-hand terms, a live player's entire year is a sample an online regular clears in a few weeks.
That gap explains a lot of self-deception at the table. If your records are a pile of buy-in and cash-out totals with no hours attached, you cannot compute an error bar or separate running bad from playing bad. A win rate without an error bar is a point estimate dressed up as a fact.
The fix is not more feel, it is arithmetic. The rest of this page works through the standard error at live speed: what the band around your number looks like, and what 100, 500 and 1,000 hours actually buy you.
Standard error at live speed
One line of maths does all the work. The width of the band around your measured win rate is set by the standard error: your standard deviation divided by the square root of your hours played. For live cash the practical unit is per hour. The win-rate confidence calculator on this site defaults to a standard deviation of 90 BB/hr, a reasonable reference for full-ring live cash. The per-100-hands figures quoted for live full ring, roughly 80 to 120 BB/100, do not map neatly onto hours, which is exactly why the live standard is hourly units. Measure your own from session logs when you can; borrow the reference until then.
The 95% band is your measured rate plus or minus 1.96 standard errors. The square root is the cruel part. The band narrows with the root of your hours, so halving its width costs four times the sample. The confidence you gain going from 100 to 400 hours takes 400 to 1,600 hours to gain again.
Worked example: 300 hours at 2/5
Say you have logged 300 hours at 2/5 NL with a £5 big blind and you are up £7,500. Step through it.
Win rate: £7,500 over 300 hours is £25/hr. At a £5 big blind, that is +5 BB/hr, a strong rate. Standard error: 90 divided by the square root of 300. The root of 300 is 17.3, so the standard error is 90 / 17.3 = 5.2 BB/hr. The band: 5 plus or minus 1.96 x 5.2, which is 5 ± 10.2. Your true rate plausibly sits anywhere from -5.2 to +15.2 BB/hr.
In money, that band runs from about -£26/hr to +£76/hr. After 300 hours of winning poker, the maths cannot rule out that you are a long-run loser who ran hot. That is not pessimism; it is what a 300-hour sample buys at live speed. The win-rate confidence calculator runs this exact sum on your own figures.
What 100, 500 and 1,000 hours buy
For the table below, hold the observed rate at +5 big blinds per hour and assume a one-hour standard deviation of 90 big blinds. The intervals narrow as comparable recorded hours increase. An interval crossing zero does not establish a positive mean under this model; an interval above zero does not prove the edge will persist.
In this fixed illustrative model, the lower bound exceeds zero once recorded hours exceed (1.96 × 90 ÷ 5) squared, or 1,244.68 hours. At an observed +8 big blinds per hour the corresponding figure is 486.20 hours. These are algebraic thresholds for selected inputs, not recommended play targets or proof of skill.
Decide off the bottom of the band
The band is not a humility exercise. It changes real decisions, because every downstream number keys off the win rate you feed it.
Bankroll is the clearest case. The risk-of-ruin maths behind this site's calculators says a true 5 BB/hr player with a 90 BB/hr standard deviation needs about 2,430 BB for a 5% risk of ruin, roughly 24 buy-ins at 100 BB deep. Feed in 2.5 BB/hr instead and the requirement doubles to about 4,850 BB, near 49 buy-ins. Halve the edge, double the roll. While your band still spans both of those figures, the honest bankroll plan uses the low end, not the headline.
The positive-drift model also has a characteristic drawdown scale: standard deviation squared over twice the win rate. At compatible inputs of 5 and 90 this scale is 810 BB; at 2.5 and 90 it is 1,620 BB. This is not an expected career maximum or a limit on losses. Maximum drawdown also depends on the horizon and path, and a real player's edge may change.
The same logic governs shot-taking. Run the move-up calculator with the bottom third of your band rather than your best guess. If the shot only works at your best-case number, it is not a shot, it is a hope.
Log so the maths can run
Retain buy-ins, cash-out, duration and stake so that results have a consistent time and money basis. Estimating an hourly variance from sessions of unequal length needs an appropriate method; the ordinary standard deviation of session totals is not automatically an hourly standard deviation. Stable inputs and complete records matter more than a borrowed benchmark.
Overshove records sessions and results. Its separate free win-rate confidence calculator can explore a model-based band from stated inputs; in-app variance intelligence is in development. A spreadsheet can also support the arithmetic if the underlying records and units are consistent. Keep the assumptions alongside any calculated interval.
Milestones, and one trap
There is no hour count that establishes an edge for every player. Read the interval for the actual result and variability, check whether the game mix is stable and keep costs consistent. Five hundred hours can still leave a wide interval; a thousand hours is not automatic qualification for a higher stake. The table above is one model scenario, not a universal timetable.
One trap on the way: do not slice a thin sample thinner. A 500-hour log split across two stakes and three card rooms produces sub-samples of 100-odd hours each, and every one of them is noise on its own. Read the aggregate first, and only trust a split when the split itself has real hours behind it.
For a fixed model and unchanged variability, more comparable recorded hours narrow the interval. Lower variability also changes its width, while selective logging or a changing game can make the model inappropriate. Do not play extra hours merely to reach a statistical threshold. Record the sessions you choose to play within your limits.
Poker win rate: BB/hr, cash/hr and BB/100
Your win rate is the long-run rate at which you win. It only means something next to the sample behind it, a figure over 40 hours is noise, the same figure over 600 hours is a read.
The three ways to express it
BB/hr (big blinds won per hour) is the live-cash standard because it compares across stakes. Cash per hour is the same edge in money, useful for life planning, useless for comparing a 1/2 game to a 5/10 game. BB/100 (per hundred hands) is the online standard, where hand counts are exact; live, you cannot reliably count hands, so BB/hr wins.
Why the number alone lies
A win rate is an estimate, and every estimate has error bars. Over a small sample those bars are enormous, a true 5 BB/hr winner can show anything from heavy losses to 15 BB/hr across a few hundred hours. Always read a win rate next to its hours, and treat short-sample figures as provisional.
What the rate includes
State the game, stake, hours and cost policy alongside a measured rate. Cash-out minus all buy-ins already reflects rake taken from pots and tips paid from the playing stack. Subtract only additional costs not already included, then label the result. No single BB/hr target proves an edge across different games and samples.
How to study live cash
Live poker has no auto hand history, so study starts with capture. This is a practical method you can use with separate notes today; Overshove's dedicated study and leak modules are in development.
Capture beats memory
You will forget many live hands. Use a brief shorthand note at the table and write the read later. A short note made near the decision is worth more than trying to reconstruct a session from memory.
Turn hands into leaks
Group your marked hands and look for patterns: over-folding rivers, barrelling turns you should give up, 3-bet pots out of position. A leak is a repeated, costly mistake, naming and tagging it is the first step to fixing it.
One action, every review
Do not leave a study session with a vague resolve to play better. Leave with one concrete change for the next session, ideally backed by a solver spot or a clear principle. Small, specific changes compound; broad intentions evaporate.
What to track in a poker session
Log too little and your stats are wrong; log too much and you stop logging. The trick is capturing the few fields that drive every higher stat, fast enough to do at the table.
The non-negotiables
Venue, stake, buy-in, cash-out and hours. From these come your net result, your hourly and your win rate by stake. Skip cash-out accuracy or hours and every downstream number is quietly wrong, this is how spreadsheets corrupt a sample.
The cost fields
Rake taken from pots and tips paid from your stack already reduce cash-out. Record additional food, travel or separately paid fees when relevant, but deduct each cost only once. A separate breakdown can explain costs without creating another deduction. State whether hours mean playing time or include travel.
The review fields
Keep a one-line hand note and a study theme in a separate note. These are optional for session statistics but useful for review. Overshove's dedicated hand and leak tools are in development.
Poker leaks: find and fix them
A leak is a repeated error, technical, mental or selection, that drains your win rate a little at a time. Big losses are memorable; leaks are expensive precisely because they are not.
The usual suspects
In live cash, the recurring ones are over-folding rivers to polarised bets, barrelling turns you should give up, over-defending out of position in 3-bet pots, limping behind in late position, and the mental leaks, tilt and the fatigue C-game late in long sessions.
Surface them from your hands
You cannot fix what you cannot name. Tag your marked hands by mistake type, then look at which tag shows up most and costs the most. A leak that appears across hundreds of hands is worth more study time than a flashy one-off cooler.
Close them deliberately
Take one leak at a time. Study the correct line, a principle or a solver spot, then make a single, specific change for your next session and watch whether the tag stops appearing. Trying to fix everything at once fixes nothing.
Win rate by stake and venue
One blended win rate averages your soft Friday game with the tough weeknight reg battle. Splitting it by stake and venue is where game selection stops being a feeling and becomes a number.
The average hides the answer
A single 4 BB/hr figure can be a 9 BB/hr soft game dragged down by a break-even tough one. Until you split it, you cannot see which rooms and stakes actually pay, so you keep sitting in games that quietly cost you.
Split by stake, then venue
Look at net and BB/hr by stake first, then by venue within each stake. The pattern usually points straight at your best game selection: the room, the night and the stake where your edge is largest.
Act on it
Once you can see your softest games, weight your hours towards them. Game selection is often the cheapest win-rate gain available, and it only appears once your log is split rather than blended.
The post-session review
You remember three hands out of three hundred, and not for long. A short cool-down right after a session captures what matters while it is fresh, and turns the session into one concrete thing to work on.
Capture while it is fresh
Within an hour of finishing, log the marked hands and write the read you could not write at the table. A note attached to a real hand now is worth more than an hour spent trying to reconstruct a session from memory next week.
Tag the leaks, not the bad beats
Separate coolers and variance from repeated mistakes. The mistakes are your leaks, and tagging them by type is what lets you later see which one actually costs the most across many sessions.
Leave with one action
End the review with a single, specific thing to change next session, ideally backed by a principle or a solver spot. One concrete change compounds; a vague resolve to play better evaporates by the next buy-in.
How much should you study
Players ask for a study-to-play ratio, but the honest answer is that quality matters more than the ratio. A small amount of deliberate study, aimed at your real leaks, beats hours of aimless solver browsing.
Deliberate beats more
Studying the spots you actually misplay moves your win rate; running random solver sims does not. Let your tagged leaks decide what you study, so every block of study time is pointed at a real, recurring mistake.
Match study to volume and stakes
A high-volume grinder needs steady, regular study to keep the edge; a lower-volume player can do less but should still review consistently. As you move up and the field toughens, the study share usually has to rise to hold the same edge.
Make it a loop, not a binge
A short review after most sessions beats a rare marathon. Consistency is what turns study into a habit and a habit into a measurable change in your leak tags over time.
Player notes that compound
Online the pool is endless and anonymous. Live, you sit with the same regs every week, and that is an edge, but only if you write the reads down instead of re-learning them each session.
A read you do not record is a read you re-buy
You work out that seat seven never folds an overpair, then forget by next Friday and pay him off again. Capturing that read once turns a single session of observation into a permanent edge against a recurring opponent.
Structure beats scribbles
Tag reads by venue and seat, with a short profile and a tell or two. Searchable and organised by room beats a pile of loose notes you cannot find when you sit down. When you arrive, the regs you have profiled are right there.
Keep it honest and private
Note tendencies, not insults, and keep the notebook private to you. Reads age, so update them when a player clearly changes, and weight recent observations over a profile from a year ago.
- →BB/hr is the live-cash standard, but it only means something next to the hours behind it.
- →Complete session fields, tagged hands and honest reviews turn results into evidence.
- →Use the lower end of the confidence band for bankroll and move-up decisions.