POKER LEAK FINDER

Find the Poker Leaks Actually Costing You EV

A leak is a recurring decision pattern with enough evidence to act on — not a red number in a HUD.

Not every deviation is a leak.

Find my leaks

REQUEST EARLY ACCESS · ALPHA IN PREPARATION

See sample report

Post-session analysis only. No real-time assistance.

WHAT IS A POKER LEAK?

A leak is a repeating decision pattern worth acting on.

A leak is a recurring decision pattern with enough evidence to act on — not a red number in a HUD.

A useful leak has four parts: a repeating decision, enough opportunities to trust the pattern, enough estimated impact to matter, and enough confidence to spend study time on it.

A LEAK IS NOT

  • An unusual tracker stat
  • One losing hand
  • A short downswing
  • Any difference from a population average

A USEFUL LEAK IS

  • A recurring decision pattern
  • Enough evidence
  • Enough estimated impact
  • Enough confidence to act
A STRANGE STAT ISN'T AUTOMATICALLY A LEAK

A strange number tells you where to look. It does not verify a leak.

Stats are triage signals. Two players can share a frequency and still make very different decisions in the hands underneath it.

SIGNAL

An aggregate frequency looks unusual.

That is a candidate area. It is not a verdict, a solver output, or a confirmed leak.

STILL NEEDS EVIDENCE

  • PositionWhere the decision happened.
  • Hand classWhich holdings created the frequency.
  • Opponent actionWhat the hero was responding to.
  • Stack depthWhether the spot was shallow or deep.
  • StreetPreflop, flop, turn, or river.
  • Pot typeSingle-raised, 3-bet, or other.
  • Decision mixWhich actual choices produced the number.
HOW WE FIND CANDIDATE LEAKS

Candidates are places to inspect, not confirmed leaks.

Stats and filters narrow the database to repeating spots. That step produces candidates. Evidence from actual hands is required before a candidate becomes a priority.

  1. SIGNAL
  2. FILTER
  3. REPEATING PATTERN
  4. EVIDENCE
  1. SIGNALA frequency or result looks worth inspecting.
  2. FILTERThe database is sliced by position, street, and pot type.
  3. REPEATING PATTERNThe same decision family keeps showing up.
  4. EVIDENCEHands are checked before anything is ranked.
HOW ACTUAL HANDS VERIFY THE PATTERN

The hands underneath the filter are the evidence.

A candidate becomes useful only when representative decisions repeat. Opportunity count and relevant decisions belong to that family — not to the whole database.

EXAMPLE

RELEVANT DECISIONS

186

of 412 opportunities

CONFIDENCE

HIGH

for this decision family

hh-river-02

CO · 3-bet pot

Called river after a missed draw

preflop / flop / turn / river

Ace-high with no improvement after a passive turn — the same branch, a different runout.

FREQUENCY ISN'T THE SAME AS COST

The more frequent pattern is not automatically the larger problem.

Frequency is one input. Estimated impact and confidence also matter. This comparison is an illustrative report output, not a universal ranking formula.

MORE FREQUENTEXAMPLE

#2

BB vs BTN Defence

−0.31

940 relevant decisions
2,104 opportunities · HIGH confidence

More relevant decisions

HIGHER ESTIMATED IMPACTEXAMPLE

#1

River Bluff Catching

−0.47

186 relevant decisions
412 opportunities · HIGH confidence

Ranked first in the sample

Prioritization can consider recurrence, estimated impact, confidence, and whether the spot is trainable. The product model may evolve. It is not a single frozen equation.

WHY CONFIDENCE MATTERS

A pattern needs enough opportunities before it deserves a strong claim.

Confidence is about the decision family being inspected. A large database can still leave a narrow turn or river spot under-evidenced.

Total hands are context. Evidence is judged on the relevant opportunities.

EXAMPLE
HIGH CONFIDENCEHigh confidence

412 opportunities

Enough opportunities in this family to support a claim.

INSUFFICIENT EVIDENCE37 OPPS

37 opportunities

Too few opportunities to verify a leak.

RANK LEAKS BY EXPECTED EV IMPACT

The first study target is the pattern with enough evidence and estimated cost.

Expected impact here is an estimated cost over the sample — not a solver-perfect EV for every node. Ranking also depends on confidence and whether the spot can be trained.

PRIORITY #1HIGH CONFIDENCE

River Bluff Catching

Calling frequency stays high versus large river bets after the hero has missed.

ESTIMATED IMPACT
−0.47
bb/100
RELEVANT DECISIONS
186
of 412 opportunities
CONFIDENCE
HIGH

The sample shows a repeating tendency to continue on the river in spots where the opponent's value range is concentrated. The finding is ranked first because the estimated cost and the opportunity count are both material.

THREE POSSIBLE OUTCOMES

Finding a candidate is not the same as verifying a leak.

EXAMPLE

A leak finder that cannot refuse a claim is just a warning generator. No action and insufficient evidence are first-class results.

PRIORITY

River Bluff Catching

Repeating pattern, enough evidence, material estimated impact.

WORK ON THIS

NO ACTION

UTG open frequency

Detected, then left off this cycle because the expected impact is small.

NO ACTION

INSUFFICIENT EVIDENCE37 OPPS

Turn probes in position

The opportunity count is too small to support a leak claim.

INSUFFICIENT
EVIDENCE

FROM LEAK DETECTION TO TRAINING

The product should produce a short study list, not a wall of warnings.

After a leak is verified and prioritized, the next step is targeted practice on that decision family — then more hands, then the same filter again.

Training on this site is the intended next step after a verified priority. This page is not a live training engine and does not advise current hands.

  1. FIND
  2. VERIFY
  3. PRIORITIZE
  4. TRAIN
  5. PLAY
  6. RE-MEASURE
RE-MEASUREMENT

Prove whether the targeted decisions changed.

The measured object is the targeted decision-pattern frequency — not a short-term winrate swing, and not a claim that training caused the result.

EXAMPLE
BEFORE
18.4%

targeted overcall frequency

6,284 NEW HANDS
AFTER
11.1%

targeted overcall frequency

CHANGE
−39.7%
STATUS
IMPROVING
CONFIDENCE: MEDIUM

Not enough new evidence to mark the leak resolved.

METHODOLOGY

Evidence before claims.

  • Evidence before claims.
  • Post-session analysis only.
  • No real-time assistance.
  • AI explains structured evidence. It does not invent hands, samples, or solver outputs.
QUESTIONS

Common questions

Is a strange stat automatically a leak?

No. Stats are signals. A leak requires a repeating decision pattern, enough opportunities, and a reason to believe fixing it would matter.

Why rank by expected impact instead of frequency?

A common small mistake and a rare expensive river error are not the same problem. Frequency is not cost.

Find your highest-priority leak.
Then train the few decisions that matter.

Find the few decisions worth fixing next.

Find my leaks

REQUEST EARLY ACCESS · ALPHA IN PREPARATION

See sample report

Post-session analysis only. No real-time assistance.