Optimal Stopping Rules in Card Drawing Games

Card drawing stopping problems are solved backward from the last draw, and the same logic prices working an order over time.

Optimal stopping problems in card drawing games ask you to decide when to stop drawing and lock in a value, rather than to compute a single expected value up front. The interview version usually strips a full deck down to a small set of cards with known face values, and the task is to state a stopping rule rather than a single number, since the rule is what proves you understand the structure.

A typical version: a deck holds cards numbered 1 through 10, one of each, shuffled and drawn one at a time without replacement. After each draw you may either stop and keep the value shown, or discard it and draw again, and you may not go back to a card you have already discarded. The question is which values you should stop on given how many cards remain.

This differs from a simple expected value question because the correct action depends on the state of the game, meaning how many cards remain and what has already been seen, rather than being fixed in advance. Interviewers are checking whether you reason backward from the end of the game rather than forward from the start, since forward reasoning tends to produce a fixed threshold that ignores how the remaining deck has changed.

With one card left, there is no decision: you take whatever value is drawn, since there is no future draw to compare it against. With two cards left, you should stop on the first card only if it is above the average value of the remaining unseen cards, because drawing again replaces a known number with the expected value of what is left.

For the ten card deck with one card already removed, the remaining nine cards average to whatever the sum of the untouched values divided by nine works out to. If the cards seen so far were low, the average of what remains is pulled higher, and the correct threshold to keep the first card rises with it. Working this out one step from the end and then extending it backward, rather than trying to solve the whole ten card game in one pass, is the tractable way to answer under time pressure.