A classic card-guessing puzzle works like this: a deck of cards is shuffled, and you try to guess each card one at a time. After every guess, you are told the true card before moving on. The goal is to maximize the total number of correct guesses on average. The interesting question is not whether you can win, but what guessing strategy is best. Three mathematicians, Diaconis, Fulman, and Holmes, proposed a specific strategy for a particular kind of shuffle called a "shelf shuffle," where the deck is divided into piles and reassembled. Their conjectured best strategy is called "direction tracking": start by guessing the lowest card, and after each revealed card, continue guessing in the same direction (up or down in value) as the most recent pair of cards, always picking the nearest remaining card in that direction.
The paper proves this conjecture and goes further. The authors show the result holds for a broad family of probability distributions over shuffled decks, not just the shelf-shuffle case. Specifically, the result applies whenever the probability of any particular arrangement of the deck depends only on how many "valleys" that arrangement has. A valley is a position where the card is lower than both its neighbors. The family of distributions covered includes any case where arrangements with fewer valleys are at least as likely as those with more valleys. The shelf-shuffle distribution turns out to be a special case of this family, which is why the conjecture follows as a direct consequence.
The mathematical core of the proof is a clever local argument. To show that direction tracking is always the best next guess, the authors compare two candidate guesses by swapping two adjacent unguessed values in the remaining deck. They show that this swap either keeps the number of valleys the same or increases it by one. Because the weights favor fewer valleys, this swap operation consistently makes the direction-tracking candidate more likely to be correct than any alternative. By applying this argument at every step and for every possible history of revealed cards, the authors establish that direction tracking is the globally optimal strategy, not just a good heuristic.