Poker Leak Finding: A Database Review Routine That Works

What Poker Leak Finding Actually Means

Poker leak finding is the process of using your own hand histories to identify recurring, costly mistakes rather than guessing at them. You filter your tracking database by position, action and board type, compare results against expectations, isolate the patterns that lose money, and convert each finding into a specific drill you practice until the habit changes.

Almost every losing or break-even player already knows something is wrong. What they lack is a repeatable method for finding out what. Reviewing hands you remember is worse than useless, because memory selects for dramatic pots and bad beats, not for the quiet, systematic errors that actually drain a win rate. A database does not have that bias.

Set Up the Database Before You Judge Yourself

Before drawing a single conclusion, make sure the data can support it.

• Sample size. Preflop tendencies stabilize quickly — a few thousand hands is enough to see whether you open too wide from early position. Postflop win rates by position need far more, and river-specific stats need more still. Treat anything under roughly 10,000 hands as a hypothesis generator, not a verdict.
• Separate your formats. Do not mix cash and tournament hands, or 6-max and full ring, or different stake levels, in one report. The correct frequencies differ enough that a blended number is meaningless.
• Use non-showdown and showdown winnings separately. The split between the two is the single most diagnostic pair of lines in any tracking software. It tells you whether you are losing by giving up too much without a showdown, or by paying off too often at one.
• Import everything. Sessions you deleted because they went badly are exactly the sessions that contain your leaks.

Which Stats to Filter On

The preflop layer

Start with the numbers that are cheap to fix: VPIP, PFR, the gap between them, three-bet frequency by position, fold to three-bet, and blind defense. A VPIP-PFR gap that is too wide usually means you are cold-calling too many hands out of position. A very low fold-to-three-bet suggests you defend your opens far past the point of profitability. These are structural, they apply to thousands of hands, and correcting them requires no new skill — only a better opening chart.

The flop layer

Look at continuation bet frequency split by whether you were in or out of position, fold to c-bet, and check-raise frequency. The most common finding here is a c-bet percentage that barely moves between dry and wet boards, which means you are not adjusting to texture at all.

The turn and river layer

Filter on turn barrel frequency after a flop c-bet, fold to turn probe, river bet frequency, and — critically — the win rate of hands where you called a river bet. River calls are where marginal errors get expensive, and most players find their single largest leak somewhere on the last street.

Positional Win-Rate Analysis

Build a report of win rate in big blinds per 100 hands, broken out by seat. The expected shape is well known: the button should be your most profitable position by a clear margin, the cutoff next, and both blinds should lose money — the goal there is losing less, not winning.

Deviations from that shape point straight at the problem.

• Button win rate that is only slightly above the cutoff usually means you are not opening or stealing widely enough in position.
• Big blind losses much worse than roughly half a big blind per hand point to over-folding preflop, or to defending wide and then surrendering too often on the flop.
• Early position win rates that are strongly negative almost always trace to an opening range that is too loose rather than to postflop play.

Do the same exercise for stack depth in tournaments. Many players have a perfectly reasonable deep-stack game and a badly broken 15-to-25 big blind game, and a blended win rate hides that completely.

Red-Flag Patterns Worth Hunting

Once the broad numbers are in view, run targeted filters for the patterns that reliably cost money.

• Calling three-bets out of position with hands that flop poorly. Filter for hands where you called a three-bet from the blinds and check the aggregate result. It is frequently the worst line in the entire database.
• River calls with bluff-catchers. Filter for hands where you called the river with one pair. If that filter shows a large loss, your calling threshold is too loose.
• Missed value on the river. Filter for hands where you had two pair or better on the river and checked. Every one of those is a bet you did not make.
• Multi-way pots. Many players' single biggest hole is treating three-way and four-way pots like heads-up pots, c-betting far too often.
• Session-tail hands. Filter the last 30 minutes of your longest sessions. If results there are noticeably worse, the leak is not technical at all — it belongs in mental game work and session-length discipline.

From Notes to Drills: Building the Study Loop

Structured note-taking

A review is only useful if it produces something you can act on. Keep one running document with a fixed format for each finding: the filter you ran, the sample size, the number that looked wrong, your hypothesis about why, and the correction. Date every entry. Never write a note like "playing too loose" — write "cold-called 47 hands from the small blind versus cutoff opens over 12,000 hands, losing at a heavy rate; replace most of that range with three-bet or fold."

Turning findings into drills

Each finding should become a repeatable exercise, not a resolution.

• Rebuild the specific preflop range in question and quiz yourself on it until recall is instant.
• Take 20 saved hands from the leaking filter and replay them one at a time, writing your intended action before revealing what you did.
• Run the spot in a solver, compare frequencies, then say in one sentence why the correct action differs from yours.
• Set a single in-session rule for the next 5,000 hands — one rule, not five — and re-run the same filter afterward to confirm the number moved.

The point of the loop is measurement. A leak is fixed when the filter that exposed it produces a different number, not when you feel better about the spot. Pair this with a clear view of your GTO strategy baseline so you know whether a deviation is a leak or a deliberate adjustment.

Make It a Habit, Not an Emergency

The players who improve fastest are not the ones who review the most hands; they are the ones who review on a schedule. A workable cadence is 20 minutes after each session marking hands, one hour weekly running filters and writing findings, and one longer monthly session comparing positional win rates against the previous month. That is roughly three hours a week, and it will outperform ten hours of unfocused hand-history scrolling.

Protect the process with sound bankroll management so a downswing never forces you to abandon the routine, and make sure you are putting in volume somewhere with reliable hand histories — compare your options among the best online poker sites before committing to a room. For guided breakdowns of the spots your filters keep flagging, our poker training videos are built around exactly these situations.

Frequently Asked Questions

How many hands do I need before database review is meaningful?

Preflop tendencies become readable within a few thousand hands. Postflop frequencies need tens of thousands, and positional win rates need more still. Use small samples to generate questions, not conclusions.

What is the fastest leak to find and fix?

Preflop range errors. They apply to the largest number of hands, they are visible in minutes with basic filters, and fixing them requires memorization rather than new judgment.

Should I use a solver for leak finding?

A solver is excellent for answering a specific question your database has already raised. It is a poor place to start, because without a filter telling you which spot matters, you will study situations that rarely occur in your games.

How do I know a leak is actually fixed?

Re-run the exact filter that exposed it after a meaningful new sample. If the frequency or the result has moved in the intended direction, the change is real. If it has not, the drill did not transfer to the table.