leak finder
Poker leak finder.
You remember the dramatic hands. The money leaves through the boring decisions you make a hundred times a week without noticing. A leak finder reads the whole history instead of the hands you happen to recall, and names the patterns that repeat. Here is exactly what it looks for, what has to be true before it says anything, and where it stays quiet.
the detectors
Four families, each with its own minimum hand count and its own thresholds. Severity is low, medium or high depending on how far past the threshold you are, and it is not a judgement of you: it is a distance.
Position
cash and tournamentA position where your result falls short of what that seat should return.
- · At least 50 hands from that position
- · A gap of at least 25 bb/100 against the reference for that seat and table size
- · The favourable end of your 95 percent interval must still sit below the reference
Priced, in big blinds per 100 hands and in euros.
Preflop frequencies
cashVPIP above the table norm plus 4, 3-bet below half the norm, or too many hands reaching showdown.
- · VPIP: at least 200 hands, above 26 percent at a six handed table
- · 3-bet: at least 400 hands, below 4 percent at a six handed table
- · Went to showdown: at least 100 flops, above 38 percent
Not priced. The loss is spread across thousands of small decisions and cannot honestly be assigned to individual hands.
Playing style
cashLimping, defending the big blind too wide, going passive after the flop, and giving up after betting it.
- · Limping: at least 80 voluntary hands outside the blinds, over 12 percent of them calls without a raise, and those hands lost money on balance
- · Big blind: at least 60 hands in that seat, over 42 percent of them seeing a flop, and defending did worse than folding would have
- · Passive after the flop: at least 80 flops, and you bet or raised in under 45 percent of them
- · Giving up: at least 25 hands where you bet the flop and saw a turn, and you bet again in under 38 percent
The first two are priced from the measured loss. The last two are not, on purpose.
Push or fold
tournamentShort stack all in decisions that disagree with the Nash and ICM reference, in either direction.
- · At least 50 graded push or fold decisions
- · Accuracy below 80 percent, with the dominant error named as too loose or too tight
Not priced. It is an accuracy measure, not a realised result.
The thresholds above are the six handed numbers. They are derived from a norm that scales with table size, so a full ring player is measured against 16 percent VPIP rather than 22, and reads their own table size in the description instead of a table they never sat at. Tournaments have their own frequency detector, because a 12 big blind decision is not a 100 big blind decision. What limping, position and variance mean is in the glossary.
four rules it holds itself to
Two of these were added after the detectors were already running, because the first runs on real data produced findings that the same data contradicted.
A pattern is only a leak if it demonstrably costs money
Wherever the money can actually be traced to the hands in question, frequency alone is not enough to raise a leak. Defending the big blind 51 percent of the time is wide by the usual standard, and the first real run of the detector reported it as a leak. The hands said otherwise: over 94 defended hands the account lost 18.72 euro, while folding all 94 would have cost 32.00 euro. Defending was 13.28 euro better than the alternative. Since 2 August 2026 both detectors where the amount is measurable require the frequency and a negative result. If the money is positive, nothing is reported.
A leak has to survive its own margin of error
The position detector used to fire on a 25 bb/100 gap alone. Over 211 hands from the button, one account measured minus 115.5 bb/100 with a 95 percent margin of 164.1, which is an interval running from minus 279.7 to plus 48.6. That interval contains zero, so the honest reading is that nothing had been measured yet. Since 29 August 2026 the detector also requires the favourable end of your interval to sit below the reference, and the margin is printed in the description so a leak that does fire cannot read more certain than it is.
The price is per 100 hands, not per month
A monthly amount depends on how much you play, so a player who took a month off would see a cost of zero on a leak that is still there. The price is stored in big blinds per 100 hands, which is stake independent, and converted to euros using your own big blind weighted by the volume you actually played at each stake. If you move up a limit, the rate stays valid and only the conversion changes.
Blank is not zero
Several detectors carry no price at all. That is a deliberate blank, not a bug and not a rounding to zero. Where the loss sits in pots you never entered, or in an accuracy figure rather than a realised result, there is no honest way to put a number on it, so none is shown.
what it cannot do
Six limits that do not go away with more development. They are properties of the data, not of the current version.
- · It only knows the hands you import. A leak that lives in the sessions you did not upload is invisible.
- · It cannot see your opponents. GGPoker encrypts their names and rotates the codes between sessions, so no detector can tell a soft table from a hard one.
- · The reference values are conventions, not measurements on this data. They scale with table size, from heads up through six handed to full ring, but they are common reference points rather than something derived from a population of players.
- · Below the minimum hand counts nothing fires. On 150 hands the report is empty, and that is the correct answer rather than a disappointing one.
- · The report shows at most five open leaks at once, ordered by price, with the unpriced ones last. Fixing the top one may simply promote the next.
- · A leak can disappear because you changed your play or because you played fewer hands from that spot. The detectors mark a leak resolved when the condition stops holding; they do not claim credit for why.
The score that the leak count feeds into, and the rest of its formula, is on the Poker IQ page.
next
See a report before you make one.
A list of thresholds is not the same as an answer. The example report shows the actual output on one account of 1,153 cash hands, with the margins next to every number, including the finding that had to be withdrawn once the margin was applied.
questions
What is a poker leak finder?
A tool that reads your own hand history and looks for recurring patterns that cost you money, rather than reviewing single hands. In this app a pattern counts as a leak only when it clears a minimum hand count, clears a threshold against a reference value, and, where the money can be traced, actually shows a loss.
How many hands do I need before it finds anything?
It depends on the leak. The style detectors need 60 to 80 hands in the relevant situation, a VPIP leak needs 200 hands, and a 3-bet leak needs 400 because 3-bets are rare enough that a smaller sample says nothing. Below those counts the report stays empty on purpose.
Why does one leak have a price in euros and another does not?
Because only some losses can be traced to specific hands. Limping and defending the big blind can be priced: the app compares what those hands actually returned against the alternative. Being too passive after the flop cannot, since the loss lives in pots you never built. Putting a number there would be a model dressed up as a measurement.
Is a leak finder allowed on GGPoker?
Analysing your own hand history after the session is a different thing from a real time heads up display, which GGPoker does not permit. The export comes from PokerCraft, the tool inside the client itself, and everything happens after you have finished playing.
Does the leak report change my Poker IQ?
Yes, one of the five components is exactly that. Open leaks subtract from a 30 point maximum: 10 for each high, 5 for each medium, 2 for each low. Three open high leaks take that component to zero.