44 Database Review, Note-Taking & Leak-Finding
Most players believe they improve by playing more hands. They do not. They improve by playing hands, recording them, and then sitting in a quiet room afterward and forcing themselves to look honestly at what the record says. The table is where you earn or lose money; the database is where you find out why. A serious modern player treats the two as a single loop (play, capture, review, adjust), driven by the tracking software on your hard drive and the notes you keep on every opponent you meet.
By the end of this chapter you will be able to mine your own hand-history database for its most costly leaks, read your win-rate graph honestly, and convert what you find into a focused, one-leak-at-a-time study plan.
This chapter is about that loop. We will cover how to mine a hand-history database for your own leaks using the major trackers, which filters and reports carry real signal, how to read the famous red line without scaring yourself off good aggression, how to keep opponent notes that are still useful three months later, and how to convert a pile of damning evidence into a focused study plan you will actually follow. Earlier chapters taught you to read a HUD live and to read opponents at the table; this one is about the cold, unhurried analysis you do away from the felt.
A database is a sample, and a sample lies in proportion to how small it is and how badly you slice it. Everything below comes with an implicit warning label: check your sample size before you believe the number. A leak you “found” in 800 hands is probably just noise. As a rough calibration we will sharpen later, a winrate needs tens of thousands of hands before a few-bb/100 reading is trustworthy, whereas frequency stats like fold-to-c-bet (how often you fold to the preflop raiser’s flop continuation bet, or c-bet) or 3-bet% firm up in a few hundred to a thousand opportunities. When the two disagree, believe the frequency.
44.1 Why review at all
Live reads are vivid and memorable; that is exactly what makes them treacherous. You remember the river bluff that got snapped off and the cooler that busted you, because they hurt. You do not remember the four hundred small pots where you folded the turn one street too early and bled a third of a big blind each time. Memory is a highlight reel weighted toward pain and drama. The database is the full footage, unweighted, and it is the only honest witness you have to your own game.
Your biggest leaks are almost never the spectacular hands you replay in the shower. They are small, repeated, boring errors — folding the big blind too much, c-betting a touch too often, calling river too wide — that each cost a fraction of a big blind but happen thousands of times. Reviewing is the act of making the boring leaks visible.
44.2 The tools
Three categories of software matter for off-table work, and most pros run more than one.
- Trackers: PokerTracker 4 (PT4) and Hold’em Manager 3 (HM3) are the two dominant hand-history databases. They import the histories your client writes to disk, store millions of hands, drive your live HUD, and (the focus of this chapter) let you filter and report on your own play. Their feature sets are broadly equivalent; pick one and learn it deeply rather than dabbling in both.
- Solvers and trainers: GTO Wizard, PioSolver, and similar tools answer the question “what should I have done here?” once a review has told you where to look. GTO Wizard in particular bridges the two worlds: it has its own analysis mode and, increasingly, can ingest your hand histories to score them against equilibrium and flag your largest mistakes by EV.
- Equity calculators: Flopzilla, Equilab, and the equity tools built into the above, for hand-by-hand range work when you want to understand a single spot in depth.
The workflow is simple: the tracker tells you where you are leaking, and the solver tells you what the leak is and how to fix it. Running a solver with no idea what to study is how people spend two hours confirming they play the button fine while the big blind quietly loses money.
44.3 Reading your own winrate: the two lines
Every tracker can draw a graph of your results over time. The single most instructive version of that graph splits your winnings into two lines.
- The blue line (or “showdown winnings”) is the money won and lost in pots that reached showdown.
- The red line (or “non-showdown winnings”) is the money won and lost in pots that ended before showdown, i.e. pots decided by someone folding.
Your total winnings are the green line, the default headline curve in both PT4 and HM3, and they are simply the blue and red lines added together. The shape of each line is diagnostic.
A red line that slopes steadily downward is the classic signature of a winning-but-leaky player: you make money at showdown with strong hands (blue line up) but give it back in the pots where you surrender, check-folding too much, folding the big blind too often, failing to fight for pots nobody wants. Most small-stakes regulars have a sharply negative red line, and that is not automatically a disease. Some of the loss is structurally correct: you post blinds you cannot always defend, and a value-heavy style legitimately wins more at showdown than away from it.
A second structural drag is one learners almost always forget: rake. Rake is skimmed only from contested pots, and it falls hardest on the blinds and on small or marginal pots, so it pushes both your red line and your blind winrates down before strategy enters the picture at all. The lower the stakes or the higher the pool’s rake, the more of your “loss” is simply the house’s cut. (Rakeback, if your room pays it, partly offsets this, but it arrives as a separate rebate rather than in the graph.) So read every benchmark in this chapter through a rake lens. The red line is a leak only when it falls steeper than your style and your rake together can explain, when you are folding equity you should be contesting.
“My red line is negative, so I must be too passive, so I need to bluff more.” This is how losing players talk themselves into spewing. A negative red line caused by too much folding is fixed by contesting more pots, not by firing three barrels with air into players who never fold. Diagnose the cause before you prescribe aggression. Plenty of solid players have a moderately negative red line, a fat and healthy blue line, and no problem at all.
The healthiest profile for most players is a strongly rising blue line and a red line that is flat-to-gently-falling. If your red line is plunging while your blue line barely climbs, you are a “showdown merchant” who only wins when you hit: predictable, exploitable, and leaving money in every pot you concede.
There is a third line worth turning on, and it speaks to this chapter’s theme of judging decisions rather than results: the all-in EV-adjusted winrate (the “all-in adjusted winnings” line). It replaces the actual outcome of every all-in pot with its EV, the money you would win on average given the equities at the moment the chips went in. One point trips people up: the all-in-adjusted figure is not the green line. The green line is still just your total winnings (blue plus red); the adjusted figure is a separate curve you switch on in the graph settings, often drawn dashed. So do not reach for the green line expecting to see your EV. A large gap between your real results and the adjusted line measures run-good versus run-bad in big pots: real results well above the adjusted line mean you have been getting there, well below it mean you have been coolered, and neither is a leak. The adjusted line is the more honest gauge of whether a soft stretch is variance or a real problem. When the adjusted line is also sliding, the deck is not the culprit.
44.4 The filters that find leaks
A tracker’s power is in filtering: carving your millions of hands into subsets and asking what your winrate is inside each one. A leak is simply a subset where you lose money (or win far less than you should) at a meaningful sample size. Here are the slices that pay rent. Winrate is conventionally measured in bb/100 (big blinds won per 100 hands).
Positional winrate
Filter your winrate by your seat. This is the first report you should ever run, and it is brutally clarifying.
| Position | Healthy 6-max winrate (bb/100) | Reading |
|---|---|---|
| BTN | strongly positive (e.g., +20 to +40) | Your money-printing seat; should be your best |
| CO | clearly positive (e.g., +8 to +16) | Second-best |
| MP / HJ | around break-even to modestly positive (e.g., 0 to +6) | Fine |
| UTG / LJ | break-even-ish (e.g., −3 to +3) | The tightest open and typically your lowest non-blind winrate |
| SB | modestly negative (e.g., −5 to −20) | Negative is normal — you post half a blind and play out of position |
| BB | the deepest-negative seat (e.g., −10 to −30) | Negative is normal; how negative is the question |
These six rows are the full 6-max table: UTG(LJ), HJ(MP), CO, BTN, SB, BB. One piece of arithmetic trips people up: you cannot add positional bb/100 figures together to get your overall bb/100. Each row is measured per 100 hands played in that one seat, and you spend only about one-sixth of your hands in each seat, so your overall winrate is the frequency-weighted average of the six, not their sum. Take the midpoints above: roughly BTN +30, CO +12, MP/HJ +3, UTG/LJ 0, SB −12.5, BB −20. They sum to +12.5, a crushing and barely believable rate, but they average to about +12.5 ÷ 6 ≈ +2.1 bb/100, exactly the small positive overall winrate a healthy winner’s profile should produce: a fat button and cutoff carrying two break-even-ish early seats and two losing blinds. If your own six seats average out negative, you are a losing player, and the leak is wherever your numbers fall furthest below these.
The blinds lose money for everyone: you invest with no choice of hand and you play out of position. In raw bb/100 the big blind is normally your single largest losing seat, deeper than the small blind, because folding 100% of your big blinds would cost −100 bb/100 against only −50 for folding every small blind, and even with flawless play you post a full blind each orbit and defend a very wide, out-of-position range. So a well-built database usually shows the BB as the most negative position, the SB a notch above it. The useful question is not whether your big blind is negative (it is, for everyone) but whether it is more negative than it should be. A big blind running worse than roughly −30 to −40 bb/100 usually means you are over-folding preflop, giving up walks and easy steals, or folding too much to c-bets. A mediocre button winrate is a different alarm entirely: it means you are leaving the single largest edge in the game uncollected.
The core diagnostic stats
Positional winrate tells you where you bleed; a handful of summary statistics that every tracker computes tell you how. Learn these with their rough healthy 6-max reg ranges, and read them in combination: each one is a symptom, and the pairings are the diagnosis.
Preflop.
- VPIP / PFR (voluntarily put money in pot / preflop raise %): in 6-max roughly 22–26 / 18–22, with the gap mattering more than either number alone. A tight gap (say 24/21) is a disciplined raise-or-fold style; a wide gap (say 28/15) means you cold-call far too much preflop, flatting hands you should be 3-betting or folding, the hallmark of a passive, dominated-range player.
- 3-bet %: around 7–10% overall. Far below that and you are too passive, letting openers in cheap and capping your own ranges (so your calls hold no premium hands); far above it with no postflop plan is just spew.
- Fold-to-3-bet: around 50–55%. Folding much more means your opens are exploitable to a re-raise (defend more or open tighter); folding much less means you are cold-calling 3-bets too wide and out of position.
Postflop. Three numbers carry almost the entire non-showdown story:
- WWSF (won-when-saw-flop), ~44–50%. One of the most informative postflop stats: it measures how often you end up with the pot once you have seen a flop, whether by showdown or by taking it away uncontested. It correlates strongly with non-showdown aggression and is a good proxy for whether you are contesting pots. But it is not the red line. Because it counts every pot you win after the flop, including the ones you drag at showdown with the best hand, you can post a healthy WWSF on the back of showdown wins while the red line bleeds, or the reverse. So read WWSF alongside WTSD and W$SD rather than as a stand-in for non-showdown winnings: a WWSF down in the 30s, especially with a low WTSD, points to surrendering too many flops and turns, but confirm it against the actual red line before you call it a leak.
- WTSD (went-to-showdown %), ~26–30%. How often, having seen a flop, you reach showdown. Too high (mid-30s and up) means you call down too wide; too low (low 20s) means you fold too much before showdown.
- W$SD (won-money-at-showdown %), ~52–56%. How often you actually win when you do reach showdown. The dangerous combination is high WTSD with low W$SD: you get to showdown constantly but win fewer than half, the calling-station profile, paying off value bets with bluff-catchers that never get there.
- Aggression frequency, roughly 40–50% postflop. Be exact about the denominator, because two similarly named stats get confused. Aggression frequency (AFq) is (bets + raises) ÷ (bets + raises + calls + folds), the share of all your postflop actions, folds included, that are bets or raises; those folds in the denominator are why a normal value lands around 40–50% rather than higher. Do not confuse it with the Aggression Factor (AF), the (bets + raises) ÷ calls ratio, which omits folds entirely and so reads as a number like 2.5 rather than a percentage. Using AFq, chronically below ~35% is the passive “check-call and hope” profile that feeds a bad red line, while a very high value can mean over-bluffing into people who do not fold.
By position and street. The same stats sliced by seat are where leaks actually surface, above all fold-to-flop-c-bet by position, the first place to look for the most common leak in poker (over-folding the big blind, dissected in the worked example below). Keep two numbers straight. The ideal solver target for BB fold-to-flop-c-bet in single-raised pots is roughly the mid-40s aggregated across c-bet sizings: you fold a touch more against the larger geometric sizings and a good deal less against small stab bets, and the blend lands in the mid-40s. Separately, treat anything around 50% and up as a practical red flag, because once you are folding half your big blinds or more after defending preflop you are almost certainly over-folding, since the solver continues on a clear majority of single-raised flops.
Showdown vs non-showdown by line
Combine the red/green concept with line filters. Filter for hands where you were the preflop aggressor, then look at your c-bet flop winrate. Filter for hands where you called a 3-bet out of position and see the carnage. Filter fold-to-c-bet by position; a frequent culprit is folding far too much to flop c-bets in the big blind, where the population badly over-folds and the solver wants you defending a wide, sticky range.
Specific high-frequency, high-stakes spots
The pots that move your winrate most are the common ones. Prioritize by frequency times severity:
- C-bet / fold-to-c-bet: happens almost every hand you raise. A small percentage error here, multiplied by its enormous frequency, dwarfs a big error in some rare spot.
- Big blind defense: you face a raise here constantly, and over-folding is the most common single leak in poker.
- Turn play after c-betting the flop: the “second barrel or give up” decision, where a lot of red-line money lives.
- River bet/call/fold: the biggest pots, so even modest frequency errors are expensive in raw bb.
Rank your study targets by frequency × cost-per-error, not by how painful the hand felt. A 0.05 bb leak that occurs 30% of hands is worth more than a 2 bb leak that occurs once every 500 hands. Trackers let you sort spots by total bb lost; let the arithmetic, not your emotions, set the agenda.
Filters for tournament and ICM players
Cash winrate is one number; tournament play needs slicing by stack depth and stage. Filter your push/fold spots when effective stacks are under ~15 bb, and check whether your shoving and calling ranges match the equilibrium charts from the short-stack chapter. Filter hands played near the money bubble and at the final table, where, as the ICM chapter explained, correct ranges tighten dramatically and a cash-game reflex to “get it in with the best of it” becomes a real, expensive leak. Note that bb/100 is a noisier yardstick in tournaments: small samples and the all-or-nothing payout structure mean you lean harder on hand-by-hand review and solver checks than on raw winrate graphs.
44.5 A worked review
Let me walk through a realistic session of self-review the way I would actually do it, so the abstractions become concrete.
I import last week’s 9,400 hands of 6-max cash into PT4 and pull up the positional winrate report. Everything looks normal except one number: my big blind winrate is −44 bb/100 over a 1,600-hand BB sample. That is worse than the −25 to −30 I would expect for my style. The sample is modest but the gap is large, so it earns a closer look.
Notice the move I just made: I did not trust the −44, I used it only as a flag to go looking. Here is the arithmetic behind that caution. A single hand of 6-max cash has a standard deviation of roughly 100 bb per 100 hands, and the uncertainty in a winrate shrinks only with the square root of the sample. Over my 1,600-hand big-blind sample the standard error is about 100 ÷ √16 = 25 bb/100, which puts a 95% interval at roughly ±50 bb/100. If anything that is too tight, because a single seat carries higher variance than the game-wide ~100 bb/100, and the big blind highest of all (a wide range, out of position, in inflated pots), realistically ~130–160 bb/100, which widens the true interval further. My −44 is therefore statistically indistinguishable from the −25 to −30 I expected. The number alone proves nothing; it only earns a look. As working thresholds, pinning a winrate to within a few bb/100 takes tens of thousands of hands (≈50k+ and climbing), and because a positional filter divides your database by about six, each seat needs proportionally more before its number means anything.
Frequencies behave far better, which is why the example pivots to one. A percentage like fold-to-c-bet or 3-bet% is an average of yes/no events, so its per-trial variance is tiny and bounded (at most 0.25), nothing like the thousands a pot measured in big blinds carries; it converges within a few hundred to a thousand opportunities rather than tens of thousands of hands. The 58% below is built on every flop I faced a c-bet in the BB, a large and fast-stabilizing count, so I can trust it long before I could ever trust the −44. When in doubt, believe the frequency, distrust the winrate, and gather more hands before acting on either.
I filter to big blind, facing a single raise, hand reached the flop, and split the result by what happened on the flop. The c-bet-faced subset is where the loss concentrates. I add a filter: big blind, faced a flop c-bet, folded. My fold-to-flop-c-bet from the BB is 58%. The solver’s baseline sits lower: a single-raised pot wants the big blind continuing on the large majority of low and middling flops, thanks to the pot odds and the preflop investment already made, folding closer to the low-to-mid 40s against the larger c-bet sizings. The population does not sit much lower, since the field over-folds the BB too, right around where I am. That does not make 58% acceptable; it makes it a shared, exploitable leak rather than a private one. Either way, I am folding the flop far too often after defending preflop.
Now I switch from the tracker to the solver to find out which flops I am misplaying. I pull a BB-vs-BTN single-raised-pot spot in GTO Wizard and look at the big blind’s flop response. The solver defends low and middling boards aggressively, floating (calling light with the intention of taking the pot away later) and check-raising connected and paired textures, folding only the genuine air. I scroll through my actual folded hands in the tracker and the pattern jumps out: I am open-folding hands like 9♥8♥ on 7♥5♠2♦ (two overcards plus a gutshot and a backdoor flush) and A♣5♣ on Q♣7♦2♠ (an overcard plus a backdoor flush and the backdoor wheel), backdoor-laden, gutshot- or overcard-equipped holdings that the solver wants to peel or raise. I have been treating “I missed the flop” as “I fold,” ignoring backdoor equity and the price I am being laid.
The leak, stated precisely: I over-fold the flop in the big blind in single-raised pots, specifically surrendering hands with backdoor straight/flush equity and overcards that should continue. That is a sentence I can study against and measure next month.
44.6 From data to a study plan
A leak you have named but not scheduled is a leak you will have again next week. The bridge from review to improvement is a written plan, and it should be ruthlessly short.
After every review session, write down exactly one leak — the largest by frequency × cost — as a single sentence, plus one concrete corrective action and one measurable check. For the example above:
- Leak: I over-fold flop in the BB vs a single raise, especially with backdoor equity.
- Action: Drill 50 BB-defense flop spots in the solver this week; build a simple heuristic (“defend any backdoor flush draw, any gutshot, any overcard pair-outs”).
- Check: Re-pull fold-to-flop-c-bet from the BB next month; target is bringing 58% down toward the low-to-mid 40s.
One leak. Not seven. You will actually fix one.
The reason for the brutal focus is that leak lists are demoralizing and motivating in inverse proportion to their length. A player who tries to fix nine things fixes zero and quits. A player who fixes one leak per week fixes fifty per year. The database will still be there; the leaks are not going anywhere. Work the biggest one, re-measure, and only then move to the next.
A sustainable rhythm for a serious player looks like: a quick session review after each session (mark 3–5 hands you were unsure of with the client’s flag/tag feature for later), a deeper database review weekly (positional winrates, red/green line, one named leak), and a solver study block between sessions targeting that week’s leak. The point of tagging hands live is that you cannot review what you did not capture, so flag the spot the moment it confuses you, because by tomorrow you will have forgotten the runout.
44.7 Note-taking on opponents
The database tracks you; notes track them. Stats describe frequencies; notes capture what a number never will: why he did something, what he showed down, the texture of his decision.
Online. Use the client’s color-coding plus short text notes; the trackers attach notes to a player that follow them across sessions. Two principles make notes useful months later:
- Record showdowns, not impressions. “Bad player” is worthless. “Showed down 8♠5♦ after calling 3 streets from the BB, river was a brick” is gold, concrete evidence of a calling range, timeless and unambiguous.
- Record deviations from baseline, not the baseline itself. Your HUD already tells you he c-bets 65%. Your note should capture what the HUD cannot: “min-3-bets only QQ+/AK,” “open-limps then 3-bet-shoves over a raise = AA/KK twice,” “tanks then jams river = always the nuts.” Capture the exploitable pattern.
A clean color scheme (say red for aggressive regs to avoid, green for recreational players to target, yellow for nits, blue for stations) lets you orient at a fresh table in one glance before any stats populate.
Tilt-notes. Typing “donkey, runner-runner, can’t fold” after a bad beat records your emotional state, not his strategy, and it actively misleads the future you who reads it and sits down expecting a fish. Notes must be falsifiable, behavioral, and written in a neutral voice. If you cannot write it as something he did and showed, do not write it.
Live. You cannot type at the table and a tracker is not running, so the channel is memory plus a discreet phone note after the session. Live notes lean on stable identifiers: the player’s appearance, seat habits, a verbal tell, a betting-size pattern (“bets pot only with the nuts, half-pot with everything else”). Encode the things from the live-tells chapter you actually confirmed at showdown, and jot them down in the bathroom or the car, because by the next session they are gone. The same falsifiability rule applies: “older gentleman, seat 4, open-limps then calls raises with any ace, never bluffs the river” is a strategy you can exploit next Tuesday.
44.8 The discipline of self-review
All of this fails without the one ingredient no software supplies: the honesty to look at a losing graph and ask “what am I doing wrong?” instead of “how unlucky was I?” Variance, as the variance chapter detailed, gives the rationalizing mind infinite cover: any stretch of bad results can be blamed on the cards, and any leak hidden behind “I just ran bad.” The database is the antidote precisely because it does not care about your feelings, but only if you let it speak.
Review your winning sessions as hard as your losing ones. Wins hide leaks beautifully: you ran a flush into a set, got there, and never asked whether the call was correct, because the result absolved you. The disciplined player audits the decision, not the outcome. A bad call that won is still a bad call, and the database is where you catch the ones the scoreboard let you get away with.
Three habits separate players who genuinely improve from players who merely accumulate hands:
- Separate decision quality from result. When you review a hand, evaluate the choice given the information you had at the time, with the river card covered. If the decision was good and it lost, log it and move on; if the decision was bad and it won, that is the hand to study.
- Quantify, then prioritize. Sort your leaks by total bb bled, not by how much each one stung. Let the arithmetic pick your homework.
- Close the loop. A review that does not change what you do next session was entertainment, not study. Every review ends with one named leak, one action, and one future measurement, and next month you check whether the number moved.
Do this and the play-capture-review-adjust loop becomes a flywheel: each session feeds the database, the database surfaces a leak, the solver fixes it, and the next session’s numbers tell you whether it worked. That loop, run patiently for a year, will move you further than any amount of raw volume. The felt is where you find out who is winning. The database is where you find out who is learning.
44.9 Summary
- A database is your only unbiased record; review it because memory over-weights dramatic hands and ignores the small, repeated errors that cost the most.
- Read the blue (showdown) and red (non-showdown) lines together, and judge the red line against your style and your rake before calling it a leak. Use the all-in-adjusted line to separate variance from a genuine problem.
- Trust frequencies long before winrates: a positional bb/100 needs tens of thousands of hands, while a stat like fold-to-c-bet stabilizes in a few hundred opportunities.
- Prioritize study by frequency × cost-per-error. The highest-value spots are c-bet / fold-to-c-bet, big blind defense, turn barreling, and river decisions.
- End every review with exactly one named leak, one corrective action, and one future measurement, then re-check the number next month.