The Red-Black Card Stopping Game

A standard 52-card deck (26 red, 26 black) is shuffled and turned face-up one card at a time. At any point before the last card, you may say "stop" — and you win if the next card is red. You must stop before the deck runs out. Does any stopping strategy beat simply calling it on the very first card, which wins with probability 1/2?

Think about it before reading on. The surprising answer is no — and proving why not is one of the cleanest applications of a martingale-style argument you'll see in an interview.

The naive intuition, and why it's wrong

It feels like there should be an edge: if you've watched 20 red cards and only 5 black ones come off the top, surely the remaining deck is black-heavy and you should wait to bet after a red-heavy run flips the odds your way? That intuition is correct in spirit — the conditional probability of red genuinely does drift above and below 1/2 as the deck depletes unevenly. The surprising fact is that no matter how cleverly you choose when to stop based on what you've seen so far, your unconditional win probability, averaged over all possible shuffles, is still exactly 1/2.

The fairness argument

Here's the key equivalence: betting "the next card is red" is exactly the same wager as betting "a uniformly random position in the deck holds a red card." Before any cards are dealt, every one of the 52 positions is equally likely to be any of the 52 cards, by symmetry of the shuffle — position 17 is red with probability 26/52 = 1/2, exactly like position 1. Your stopping rule only decides which position you're betting on; it can't change the fact that each position, examined in isolation before the shuffle, is red with probability exactly 1/2.

More formally: let $S$ be your (possibly random, possibly history-dependent) stopping position, chosen using only information from cards already revealed. Since $S$ depends only on the past and not on the future card, and the identity of card $S+1$ is symmetric under swapping any two not-yet-revealed cards, the event "card $S+1$ is red" has probability exactly $26/52$ no matter what rule generated $S$. This is a martingale argument: the process "fraction of remaining cards that are red" is a fair game — your information updates as cards are revealed, but there's no way to convert that information into a systematic edge on the next single card, because the shuffle was uniform and every card is equally likely to occupy every remaining slot.

Worked check: a tiny deck

Take a 4-card deck: 2 red (R), 2 black (B). Verify with brute force that "always bet on card 1" and "wait, then bet after seeing exactly one card" both give exactly 1/2.

There are $\binom{4}{2} = 6$ equally likely arrangements of 2 R's and 2 B's: RRBB, RBRB, RBBR, BRRB, BRBR, BBRR.

Strategy A — bet on card 1 always. Red is card 1 in RRBB, RBRB, RBBR — 3 of 6 shuffles. Win probability $3/6 = 1/2$. ✓.

Strategy B — "stop as soon as you've seen a black card, then bet the next card is red" (skip position 1 if it's already red, since you want to catch a black-heavy stretch). Apply it card by card to each shuffle: RRBB → first card R, rule says only stop after seeing black, so wait; second card R, still wait; third card B — stop, bet card 4 is red, but card 4 is B: lose. RBRB → first card R, wait; second card B — stop, bet card 3: it's R: win. RBBR → first R, wait; second B — stop, bet card 3: it's B: lose. BRRB → first card B — stop immediately, bet card 2: it's R: win. BRBR → first card B — stop, bet card 2: it's R: win. BBRR → first card B — stop, bet card 2: it's B: lose.

Tally: win, lose, lose, win, win, lose — 3 wins out of 6, exactly 1/2 again. The "smarter" strategy, despite reacting to what it's seen, nets the identical 1/2.

The generalizable technique

The reusable move — recasting "when do I bet" as "which fixed position am I effectively betting on" — is a fairness or symmetry argument, the same family of tool used to show a fair coin's stopping rules can't beat 50/50 in a related problem, and it's the seed of why trading strategies that only use past price information can't have an edge on an efficiently random next tick without some genuine informational advantage. Whenever a puzzle asks "can cleverness beat the base rate," check first whether the underlying process is exchangeable — if every future outcome was equally likely under the original random draw, no amount of watching changes the unconditional odds.