arXiv nightly build-off ¡ arXiv:2606.29530

Cutoff: The Shuffle That Suddenly Mixes

Drag block size m and deck size n, watch a real simulated deck get riffled, and see the moment it snaps from "obviously not random" to "statistically indistinguishable from random" — not gradually, but all at once.
What "cutoff" means: most shuffling processes don't mix gradually — they stay almost maximally far from random for a long time, then collapse to random over a short window, like a switch flipping. This is the cutoff phenomenon. For the classic top-to-random shuffle (move 1 card at a time), that switch flips around n·ln(n) shuffles. The paper behind this demo, "Cutoff profiles for colored top-m-to-random shuffles with growing block size" (arXiv:2606.29530), studies what happens when you move a whole block of m cards at once instead of one — it works out the precise shape of that collapse, and shows the shape itself changes as the block grows relative to the deck.
Simulation below is a genuine Markov chain (real deck permutations, real random insertions) run over a population of parallel decks — not canned numbers. The dashed "theoretical profile" overlay is a Poissonized asymptotic approximation in the spirit of the paper's untouched-card analysis, shown to illustrate the qualitative shape, not to reproduce its proofs exactly.
Live deck bar = 1 card ¡ color = original position
Shuffles so far
0
Predicted cutoff
—
P(deck still ordered)
100%
A fresh deck. Every card is exactly where it started — deal with confidence that you know where the aces are.
Distance from random total-variation-style proxy
Empirical (300 simulated decks) Theoretical asymptotic profile Current step Predicted cutoff window