Show the math

The Lab methodology

The Lab cards on your dashboard — and the ranges next to your opening win rates — run on four rules. Same doctrine as this page: a number without its uncertainty is a guess wearing a suit.

• Win-rate ranges like “47% (41–53)” are Wilson 95% score intervals. Wilson rather than the textbook normal approximation because it stays inside 0–100% and behaves at small samples and extreme rates — which is exactly where a chess record lives.
• “Plays like” is a performance rating: your average opponent's rating plus 400 × (wins − losses) ÷ games, the standard linear form. It tells you what a raw win% hides — 50% against stronger players is a better result than 50% against weaker ones. Under 15 games it isn't shown at all; below that the number is noise dressed as precision.
• The tilt test compares your win rate in the game right after a loss against your rate after a win or a draw — two disjoint sets — with a two-sided two-proportion z-test at α = 0.05, and it needs at least 30 games on each side before it will speak. Not significant means the card doesn't render, full stop.
• Hour windows are 3-hour blocks by when your games ended, in UTC (not your local clock — we say so on the card). A block only becomes a claim with 25+ games AND its whole Wilson interval above (or below) your win rate at every other hour. We check 24 overlapping windows, so the most extreme block is often luck — the interval test is the guard.

Silence is a result. If a Lab card is missing, the pattern didn't clear the bar — we don't round a maybe up to a finding.