one life to live · metaphor 36 of 100
How long should you search before you commit — to the apartment, the hire, the person? Opportunities arrive one at a time, you must decide on the spot, and the ones you pass rarely wait. That predicament has a solved core, and its answer is a number.
Everyone knows both failure modes, usually by first name. There is the friend who married the first serious relationship and carries, twenty years on, a small locked room of wondering. And there is the friend who kept auditioning — flats, hires, partners — measuring each against a composite of all the others, until the good ones had quietly paired off with people less exacting and the searching had become the life.
Between those two mistakes the advice industry offers a platitude: trust your gut; when you know, you know. Mathematics offers something stranger and more practical — a strategy with a schedule. The problem of the sequential, irrevocable choice was solved in the early 1960s, and its solution is a number: spend the first 37% of the search committing to no one while remembering the best you meet; then take the first candidate who beats that memory. Below, twenty apartments are about to come on the market, one per day, each gone by nightfall.
the predicament
Three constraints, jointly, define the game. Opportunities arrive in sequence, not side by side: you never survey the whole field, you meet it one candidate at a time, in an order you don't control. The decision is immediate: the apartment wants a deposit by evening, the candidate is holding a competing offer, the person across the table will reasonably conclude something from years of noncommitment. And there is no recall: the ones you pass rarely wait to be reconsidered.
When all three bind, more searching is not simply more information — every look burns the thing looked at. Commit early and you choose in ignorance of what the market holds; commit late and you choose from its leavings. The question is when to stop — and "when" is exactly the kind of question that has an optimum.
the 37% rule
The optimal strategy has two verbs. First, look: let the opening stretch of the search pass entirely — you will take none of them, however lovely — while remembering the best you see. This is not window-shopping and it is not wasted time; it is calibration. You cannot know what a good apartment, hire, or partner is for you, in this market, at this price until you have seen enough of them, and the look phase is precisely the purchase of that knowledge. Its entire output is a single number: a benchmark, the best real candidate so far.
Then, leap: take the first candidate who beats the benchmark. The rule never asks whether this one matches the best imaginable, only whether it beats everyone actual you learned on. And the schedule is exact: set the look phase to 37% of the pool and your chance of landing the single best is itself about 37% — both fractions are 1/e, one of the theorem's quiet jokes. Blind luck gives 5% here, and no strategy of any kind beats the rule.
what to try
Drag k to 0% and run the playthrough: the first-love strategy. It commits instantly, before any benchmark exists, and wins exactly as often as chance — one time in twenty. Now drag it to 95%: the eternal auditioner. The benchmark climbs so high during the long look that almost nothing clears it, and the run usually ends with a forced settling for whoever happens to be last. Both fail, but they fail differently — one on ignorance, the other on depletion — and the curve in the sweep shows the whole trade at once, sagging at both ends and cresting near 37%. (Changing k replays the same market, so you can watch one sequence of candidates reward one temperament and punish another.)
Then flip on the second curve. Score a search as a success if it ends anywhere in the top 10%, not only at the very best, and the peak slides earlier and rises higher: you should stop calibrating sooner, and you will be satisfied far more often. It is perfectionism — the insistence that only the best counts — that makes the schedule long and the odds harsh.
the schedule of a search
Translate the number into a calendar. Given a year to find an apartment, roughly the first four and a half months are calibration: view, decline, take notes on what the market actually offers. Twenty interviews for one role: the first seven are for learning what a strong candidate in this pool sounds like. A search you expect to run from twenty-five to forty: the benchmark-building years end around thirty.
The rule's deepest gift is permission, extended in both directions. Permission to decline early without guilt — refusing a good candidate in the look phase is the tuition the strategy charges. And permission, once the benchmark breaks, to stop — to treat the itch for one more look as what it now is: a bet with worse odds than the one standing in front of you. The search gets a schedule precisely so the life doesn't have to become the search.
the mapping
| Mathematics | Life |
|---|---|
| the candidate sequence | Opportunities arriving one at a time, in an order you don't control. |
| no recall | Passed chances rarely wait: the flat is let, the candidate signed, the person moved on. |
| the look phase (first k) | The relationships, rentals, and interviews whose real job is to teach you the distribution. |
| the benchmark | What "good, for me, in this market" concretely means — the best real thing seen so far. |
| the leap rule | Commit to the first that beats everything you learned on — not the best imaginable. |
| k/n ≈ 37% | The calibrated share of any search to spend learning before you're allowed to choose. |
where the metaphor tears
The classical objective scores the single best candidate as 1 and everything else — including a second-best nearly indistinguishable from the first — as 0. Almost no human holds this utility about apartments or spouses. The theorem answers the perfectionist's question; relax the objective even slightly (the top-10% toggle above) and the optimal look phase moves earlier while the odds of a happy ending jump from roughly 38% to over half — and the looser the standard, the shorter the schedule gets. If the 37% rule feels punishingly long, that is the perfectionism talking, not the mathematics.
Real searches leak at both assumptions. Passed candidates occasionally remain available, apologetic phone call and all; and the fiction that the pool holds exactly twenty, or the window exactly a year, does quiet load-bearing work. Variants with recall, unknown horizons, and candidates arriving in time rather than in counts all exist, and each moves the number — mostly earlier. What survives every variant is the shape: an opening phase of deliberate noncommitment, then a threshold rule.
The theorem assumes the judge is constant: the person ranking candidate nineteen is the person who ranked candidate one. But looking is formative. After enough apartments you want different things from an apartment; after enough relationships the benchmark is not just higher but aimed somewhere else. The model treats the look phase as pure measurement, when in life it is also manufacture — the standard you leap with was partly built by the search, and it goes on drifting after you stop.