Indexing the archive…
Your Universe of Digital Possibilities
Candidates pass one at a time in random order, and you must decide on the spot — with no second chances. Commit too soon and you gamble before you've learned anything about the field; wait too long and the best candidate has already walked past, unrecognised or already gone. The optimal balance is to watch the first 1/e of them purely to calibrate, then pounce on the next record — and it wins the very best about 37% of the time, no matter how many candidates there are.
Candidates arrive one at a time in random order; you must accept or reject each on the spot, and you want the single best. The rule: watch the first r ≈ N/e to calibrate, then grab the next one that beats everyone so far.
The chance the look-then-leap rule lands the very best, as a function of the cutoff r. Too small and you leap before you have calibrated; too large and the best has likely already walked past. The maximum sits at r/N ≈ 1/e.
The striking part: as the field grows, both the optimal cutoff fraction and the probability of success converge to the same number, 1/e ≈ 37% — whether there are ten candidates or ten million. Skill cannot push a random order past this wall.
A single rule for a whole family of stopping problems: sum the odds of the remaining “records” and stop the moment that running sum reaches one. The secretary’s 1/e is just this theorem applied to a random permutation.
Reject the first r candidates no matter how good, remembering only the best of them, then accept the next candidate that beats them all — that single rule maximises the chance of landing the very best. The optimal r sits near N/e, and there the win probability itself converges to the same constant, 1/e ≈ 37%, a limit that holds almost exactly regardless of how large the field is. It connects to The Bandit(INST·76): deciding when to stop exploring arms and settle on exploiting the best one is itself an optimal-stopping problem, just repeated. The puzzle's first telling was Merrill Flood's 1949 “fiancée problem” — the same Flood who, with Melvin Dresher, first posed the Prisoner's Dilemma explored in The Tournament (INST·33). And The Oracle(INST·24) is the sibling case of committing to decisions against a randomly unfolding future. Bruss' 2000 odds theorem folds the whole family into one line: sum the odds of success on each remaining trial, and stop the instant that running sum first exceeds one.