No peeking. Make every permitted rule testify.

The exact law of a fair prediction

For a prediction $P$ and draw $A$, let $H=|P\cap A|$. Under a fair $k$-of-$N$ draw,

$P(H=h)=\frac{\binom{k}{h}\binom{N-k}{k-h}}{\binom{N}{k}}.$

That law is identical for every rule choosing $k$ numbers using only past information. The total-hit law is the exact convolution across locked trials.

The interface exposes five strategies and fixed horizons of 50, 100, and 250 draws when history permits them. All available combinations form one registered family. If its smallest raw value is $p_{\min}$ across $M$ trials, the court reports $\min(1,Mp_{\min})$ as its whole-court verdict.

Discovery is not prediction

A pattern is easy to discover after the answer is visible. Pattern Court turns folklore into a prospective experiment: train first, lock the decision, then reveal. A good story enters evidence only after surviving that boundary.