A poker leak is a mistake you repeat, not one you made. That distinction is everything: the hand you misplayed on Tuesday cost you a pot; the pattern that produced it costs you a pot a week for a year. Which is why leaks are not found in the memory of your last session, but in the record of a thousand.
One-off mistake vs leak: frequency decides
You called a river that was an obvious fold. Leak? Not yet. It becomes one when the filter shows you call 70% of rivers in big pots out of position. A mistake is a data point; a leak is a tendency. And only tendencies get fixed with a rule.
Practical consequence: you cannot find leaks without data. A tracker database, a well-kept session tracking sheet, or both — but without numbers, what you are collecting is suspicions.
The six most expensive (with their reference number)
1. Open-limping preflop. The most common one live and at micros. Entering by calling the blind gives up the initiative and any chance of taking the pot without seeing cards. Your limp frequency should sit near 0% at a normal table. You raise or you fold.
2. Playing too many hands from early position. The reference for a healthy 6-max reg: VPIP 22 to 26, PFR 18 to 22, with a gap of 5 or less between them. If your VPIP and PFR read 35/14, you are not playing loose, you are calling loose, which is a different thing and far more expensive.
3. Auto-piloting the c-bet. Betting 90% of flops works right up until someone pays attention. On boards that hit the caller's range (T98 two-tone, for example), your c-bet frequency needs to drop and its selection needs to improve. The leak is not betting, it is betting without reading the board.
4. Never folding top pair in big pots. The classic that takes a full month of winrate in three hands. If the opponent has fired three streets and the pot is big, your top pair with a medium kicker is rarely ahead. This one is fixed by reading ranges, not guessing: hand reading is exactly that job.
5. Playing with no table selection. Sitting at the first open seat means accepting any field. Half an hour of table selection a month moves your results more than twenty hours of theory. And when rake takes a large slice of every pot, the leak stops being technical: check how much comes back to you in rakeback before blaming your game.
6. Sessions with no exit rule. Playing five hours when your concentration is good for two is a discipline leak, and usually the most expensive of all because it contaminates the other five. A stop-loss written down before you sit is not cowardice: it is the guardrail your best self leaves for your worst one.
Three methods for finding yours
The position filter. Open your database and look at bb/100 in each seat separately. Watch the trap: the big blind always loses. A winning reg loses somewhere between 25 and 45 bb/100 from the big blind, so there you compare against that benchmark, not against zero. Where the leak really shows is button and cutoff: if those two are not your most profitable seats, the problem is in how you open and steal, not in your luck.
The live-marked review. During the session, flag the hand you were unsure about — not the one you lost, the one you doubted. Afterwards, review three. Three properly reviewed beat thirty skimmed. This is the method that makes study sustainable; the rest of the routine is in how to study poker.
The second opinion. Your blind spot is blind to you by definition. A study partner reviewing your hands finds in twenty minutes what you miss in a month, because they do not have to defend your decisions.
How many hands before you conclude
Almost everyone gets this wrong. Preflop stats (VPIP, PFR, 3-bet) stabilise around 10,000–20,000 hands. Late-street stats — river fold, turn c-bet — need 50,000 or more. At 3,000 hands you do not have a leak, you have noise. Drawing conclusions from a short sample means trading your game for variance, which is the fastest way to get worse while working hard.
The golden rule: one at a time
Finding six leaks and attacking all of them on Monday ends with zero fixed. Pick the most expensive one, turn it into a one-sentence rule ("I never limp", "I fold weak top pair to the third bet"), run it for a full month, and measure. Then the next one.
The lesson
Leaks are not fixed by playing more, they are fixed by looking. Your graph is not a grade, it is a map — and the part that slopes down always has a repeated reason behind it, which is almost never the one you assumed.
To filter by position and see those numbers you need a tracker: the ones we use are on the tools page.
Which leak was the hardest for you to fix? Share it in the community and save someone months. Join free.