Skip to content

Database Review & Leak-Finding

Your tracker tells you what to study in the first place. Studying random spots is guessing; studying your actual leaks is deliberate practice. This is how you find them — and how to avoid being fooled by small samples.

Find leaks by comparing to winners, not the pool

Pull your core stats (VPIP, PFR, 3-bet, fold-to-3bet, c-bet flop/turn, fold-to-c-bet, WWSF, WTSD, W$SD, aggression) and compare them against the biggest winners at your stakenot the pool average.

The average player is a losing player, so matching pool averages is worthless. Match (and understand) what the winners do.

Stats alone lie — filter to the actual hands

A "60% turn-barrel" number means nothing without context. Use your tracker's filters (PokerTracker's Actions & Opportunities, HM3's filter editor) to pull the specific hands behind a suspicious stat — which boards, which opponents, which lines. The stat flags a maybe; the hands tell the story.

Sample size: trust opportunities, not raw hands

A stat is only as reliable as the number of times its situation came up:

Reliability Opportunities
Working minimum ~10
Solid 10–30
Trustworthy 30+

Preflop frequency stats (VPIP, PFR, RFI) stabilize fast. Turn/river stats need far more — fold-to-turn-cbet ~300+ hands, river stats ~1,000+. A flashy read on a tiny sample is noise.

Population reads (no solver needed)

Your database is a population model. Build a report across all players in your DB (grouped by stake) to get pool-wide tendencies — how often the field folds to 3-bets, calls steals, etc. These population reads are how you exploit unknown opponents, and they feed node-locking and hand reading.

The variance reality check

You cannot read your results over short samples — variance is enormous. Standard deviation in 6-max NLHE is ~90–100 bb/100, which means:

Even 100,000 hands isn't certain

At 100k hands, your true winrate is pinned only to about ±5.6 bb/100. A measured 5 bb/100 could really be anywhere from ~0 to ~11. So never change your strategy off a downswing within a normal sample — judge your decisions, not your graph. → Mental Game & Variance

(Compute your own confidence intervals with a variance calculator: 95% CI = winrate ± 1.96 × SD ÷ √(hands/100).)

Key takeaway

Diagnose before you study. Compare to winners, filter to real hands, trust 30+ opportunities (not raw hands), mine population reads — and never let a small-sample graph dictate strategy.


Sources: HUD stat sample sizes · database review method · variance