A typical self-review looks like this: open the positional report, find a stat that differs from the "norm", declare it a leak. A month later the stat is pulled toward the norm and the winrate has not moved. This happens because a stat describes a frequency, while the money leaves in specific decisions. Below is the order in which a database is worth reading so you get down to decisions instead of stopping at frequencies.
1. Sample first, conclusions second
The first question for a database: can it tell you anything at all. The spread of winrate in cash games is huge: with ordinary variance, 100 hands give a deviation of about 100 bb/100. Over distance it smooths out, but slower than it seems. Much slower than it seems after a good session.
| Hands in the database | Winrate interval, bb/100 |
|---|---|
| 50,000 | about ±9 |
| 100,000 | about ±6 |
| 225,000 | about ±4 |
| 1,000,000 | about ±2 |
This means that on 100,000 hands a winrate of +3 bb/100 is statistically indistinguishable from zero. It is too early for conclusions about the total. Conclusions about decisions, however, are possible: there are thousands of specific spots in a database and each has its own sample. So from here the review goes not from the total but from spot to spot.
2. Rake and all-in EV
Before looking for mistakes in play, look at how much leaves outside of play. At NL50 rake takes 5 bb/100 and more, at micro stakes even more. A player at +4 bb/100 before rake and −1 after rake does not "play badly", they lose because of the structure of the stake. What needs fixing there is not spots but table selection, rakeback or the stake itself. You can spend a year polishing your BB defence and feed the room the whole time.
The second adjustment: the difference between the actual result and all-in EV. If it is large, part of the minus or plus in the database is random, and spots with all-ins should be read with that in mind.
3. Preflop against the solver
Preflop is the cheapest to check: there are many decisions, the sample in each hand category fills quickly, and the solver's ranges exist. Compare your opening ranges, blind defense, 3bet and response to 3bet with solver ranges or with the ranges of the top regs at your stake.
What to look at first:
- Opening first in from late positions and from the SB. This is where you most often find either too wide a range of offsuit trash or, the opposite, folds of hands the solver almost always opens.
- Response to a 3bet: hand categories that systematically do worse than a plain fold, and categories played less often than by the solver.
- BB defense against steals: not the overall frequency but which hands fold against which size.
A preflop mistake rarely costs much on its own, but it repeats thousands of times and drags the whole postflop along with it. Opening Q7o from the CO is cheap. Playing the flop, turn and river with it is not.
4. Red line and blue line by position
Now look at where exactly the money leaves: with showdown or without. The blue line, showdown winnings, is positive for most regs. The red line, non-showdown, is negative. In itself that is not a mistake: the blinds you posted and folded are written into the red line too.
It is useful to look at both lines by position and by pot type. If the red line in the BB in single-raised pots against CO and BTN is noticeably worse than for players at your stake, the spot is found, and from there it has to be opened up street by street.
5. Spots by money, not by frequency
The main mistake of self-review: take the stat that deviates the most and fix it. A deviation in points says nothing about price. A flop cbet of 75% against 60% for top regs may cost zero if on the turn and river you play like them. And a fold against a 45–55% pot bet on the river that differs by 5 points may cost bb/100. The reddest stat in the tracker and the most expensive leak are usually different things.
So the order is this: first the spot where the most money passes through your decisions. As a rule that is the single-raised pot in the BB against a late-position open and the single-raised pot where you are the aggressor from CO or BTN. Then inside the spot by street: how often you choose a line and how much it brings compared with the same lines for top regs.
6. The river separately
If you had to pick one place where regs at NL25–NL200 most often keep their most expensive mistakes, it is folding bluffcatchers on the river. The pool at these stakes bluffs less than is needed to make large bets break even, but that does not mean every bluffcatcher should be folded. The expensive folds are against medium sizes in small pots, where a call pays off even with a small share of bluffs.
Such folds must be checked by hand: by bet size, pot type and position, with a dozen real hands in front of you. A number without the hands read remains a hypothesis. And a hypothesis you have already rebuilt your game around is an expensive hypothesis.
7. What not to do
- Do not fit stats to a "norm". Every stake and pool has its own norm, and a stat may deviate because you adapted correctly.
- Do not fix everything at once. The three most expensive findings in a month give more than a list of thirty. A list of thirty usually ends with nothing fixed.
- Do not draw conclusions from spots with a sample of dozens of hands. The price of a postflop spot needs roughly a hundred times more hands than a frequency.
- Do not judge play by winrate over a short distance. Judge it by the price of decisions: it is counted over thousands of repetitions of the same spot and converges much faster.
Summary
The order of review: sample, rake and EV, preflop against the solver, lines by position, spots by money, river by hand. At every step the question is the same: does this cost money or is it just different? If there is no answer, it is not a leak yet. It is just you.