How many chess games before a pattern actually counts?
Sample size is the difference between a habit and a story. Here is how Chessigma decides a repeated opening choice is worth reporting.
A pattern counts when the same position appears often enough that both the move you played and the move you usually play have a sample above the noise floor. One loss is not a repertoire problem.
Why five games is a floor, not a conclusion
Broad splits — colour, time control, a named opening — can be described once a bucket has at least five games. That is enough to say “this slice exists,” not enough to tell you to abandon an opening.
Repeated-position findings use a higher bar. Both candidate moves need multiple games, and the score gap has to clear a gap that a pair of upsets cannot explain.
What changes the count
A rating jump, a new time control, or a month of arena chess will move your baseline. Compare recent games to recent games. Mixing a 2024 bullet binge with this month’s rapid hides the thing you are doing now.
Chessigma’s public preview reads up to 80 games; a saved import can read 200. The cap is there so the newest corpus stays the one you train on.
Frequently asked questions
Why does a report sometimes say nothing?
Because nothing cleared the threshold. That is more useful than inventing a weakness from two games.
Should I import every game I have?
No. Prefer the current time control and the last few months unless you are specifically studying an old repertoire.