How many games do you need before your stats mean anything?
Play four games in the Italian, win three, and a dashboard that speaks too soon will tell you the Italian is working. Lose the next six and the same dashboard will tell you it is a weakness. Nothing about your chess changed. The sample did.
This is the fastest way to teach someone that a stats page is noise: let it talk on four games. Chess Wizard’s insights page refuses to. The floor is 15 games for anything that is a rate over games, and 40 moves for anything that is a rate over moves. Trend — whether your accuracy is up or down — needs about 30, because it is a comparison of two halves.
Those numbers are not a ritual. They are roughly where a confident sentence stops being a coin flip.
The arithmetic, without a lecture
Suppose your true win rate in some opening is 50%. In five games the chance of a 80%+ sample is still common. In fifteen it is unusual. In forty it is a story. You do not need a statistics course to feel this: you have had weeks at the board that felt like a breakthrough and were, in retrospect, a streak.
Move-level numbers converge faster than game-level ones because a blitz game is 30–40 of your moves, not one data point. Forty moves is about one messy game or two quiet ones — enough that a 3% blunder rate in a clock bucket is not a single hang. It is still not a lot. It is the point at which we are willing to speak, not the point at which the number has become truth.
A claim like “you blunder three times more often with under 30 seconds on the clock” has to survive two floors at once: enough moves in the bad bucket, and enough in the baseline to compare against. If either side is thin, the sentence does not appear. A missing insight is not a missing feature. It is the product working.
What 15 games is not
It is not “your rating is accurate.” Rating is a different estimator, with its own floor, and a 1200 on Tuesday is not a 1200 on the same scale as a 1200 in 2019. If you want to know whether you are improving, compare your last twenty games to the twenty before that, at the same time control. That is why the trend sentence needs about thirty games total.
It is not “play 15 games of an opening and then trust the score.” Fifteen is the minimum before we will name a weakness. A 40% score over 15 games is a smell. A 40% score over 60 is a finding. The insights sentence includes the sample size on purpose: “31% over 42 games” and “31% over 15 games” should not feel the same to you, even when the headline number matches.
It is not a reason to grind volume for its own sake. Fifteen honest rapid games will tell you more than eighty bullet games you barely remember. The floor is about sample size, not about hustle.
What to do while you are under the floor
Play. Look at individual games. The eval graph, the key moments, and the move list are about this game; they do not need a population. The eval graph post is the right tool in that stretch.
When the insights page finally has something to say, believe the sentence that cites the most games, not the one with the most dramatic number. Dramatic on a small sample is how you end up rewriting a repertoire around a weekend.