the second hundred · metaphor 144
Two people choose each other and call it mutual. But one of them moved first — asked, reached out, risked the no — and that quiet fact shaped who each of them ended up with, more than either would admit.
A courtship feels symmetric from the inside. We found each other; we both said yes; it was a meeting of equals. Yet almost always one side pursued and the other side chose among those who pursued. The pursuer looks vulnerable — they can be turned down. The chooser looks powerful — they hold offers and pick. It is worth asking which role actually does better, because the answer is the reverse of the feeling.
There is an algorithm — a real procedure used to marry doctors to hospitals and children to schools — that resolves a whole crowd of such courtships into a matching nobody wants to walk away from. It always succeeds. But it is not neutral: the side that proposes gets the best partner it could possibly have in any stable outcome, and the side that waits gets the worst. The tilt is invisible inside any single couple. It is unmistakable once you can flip the roles and watch.
Deferred acceptance
Gale and Shapley described the procedure in 1962 and proved two things about it. First, it always finishes with a stable matching: there is no pair of people who both prefer each other to whom they got. If such a pair existed they would defect, and the arrangement would unravel; stability means it never has to. Second — the part that stings — the outcome it reaches is not one stable matching among many. It is a very particular one.
The rule is simple. Every person on the proposing side asks their favourite. Every person on the receiving side who has offers keeps the best one on hold and rejects the others — not forever, just for now. Anyone rejected asks the next name on their list. A held offer can still be bumped later by someone better. The music stops when no one is left holding a rejection with names still to try.
The theorem: this produces the proposer-optimal stable matching. Every proposer gets the best partner they could have in any stable matching at all — and simultaneously every receiver gets their worst. Same people, same preferences, one procedure: whoever holds the initiative walks away better off, across the whole market, every time.
What to try
The panel is a live little market — five people a side, preferences drawn fresh each time you press New market, nothing scripted. Step plays one proposal at a time: a gold line shoots out, a held line survives or a red rejection flashes, and you can watch a receiver upgrade as a better suitor arrives. Run races to the finish. Read the two chips: left avg rank and right avg rank — the average position of each person's partner on their own wish-list, where 1 means they got their first choice.
Now the whole point: hit Right proposes on the same market. The stable matching shifts. The proposing side's average rank drops toward 1 — better partners — and the other side's climbs. The proposer advantage chip measures exactly what the initiative buys: how many rank-points better the proposing side does than it would as the mere receiver (it reads ±0 · unique only on the rare market with a single stable matching). And blocking pairs stays at zero throughout: both outcomes are perfectly stable, just tilted opposite ways. Nobody's preferences changed. Only who spoke first.
The mapping
Read the sides as the two postures in any courtship. The proposers are the ones who initiate — who ask out, confess, apply, make the offer, and absorb the rejections that come with going first. The receivers wait and select among whoever shows up. The felt power is with the receiver: they hold the offers, they say yes or no. The mathematical result is that they pay for it. By never reaching past whoever happened to arrive, the receiver ends up with the least-bad of their suitors, while the proposer, willing to be turned down all the way down their list, lands the best partner the situation will stably allow.
This is why "we chose each other" hides a seam. Both statements are true — the matching is mutual, neither would defect — and yet the two of you did not do equally well relative to the field. The initiative you exercised or declined months ago is still there in the outcome, silent, structural. It generalises past romance: whoever sends the first serious signal in a job market, a negotiation, a collaboration is, other things equal, buying a better place in the final matching than the one who waits to be approached.
Read as life lessons
The side that proposes does strictly better — provably, market-wide. The vulnerability of going first is not weakness; it is the move that captures the best attainable partner. Waiting to be chosen is the expensive posture.
A matching with no blocking pair can still pair everyone with a mutual second-best. Stability only means no two people would both defect — not that anyone got what they wanted. Contentment is a different measurement.
Inside one couple you cannot see it; both said yes freely. The asymmetry lives in the counterfactual — the other stable matching you'd have reached if the roles had flipped. You only see it by comparing.
In the wild
The U.S. residency match (NRMP) marries thousands of new physicians to hospitals with a deferred-acceptance algorithm each spring — and it was redesigned in the 1990s to run applicant-proposing, on purpose, to tilt toward the doctors.
Boston and New York rebuilt their public-school assignment on the same idea, so that no family and school would both rather have swapped — and so that gaming the form stopped paying.
Kidney-exchange chains and, informally, the whole economy of dating and hiring run on who-proposes-to-whom. Roth shared a Nobel in 2012 for turning this into working market design.
The mapping, exactly
| Mathematics | Life |
|---|---|
| the proposing side | Whoever initiates — asks out, applies, makes the offer — and accepts the risk of being turned down going first. |
| holding an offer | Keeping your best current option while entertaining better ones — the posture of the pursued. |
| a stable matching | A pairing nobody defects from: no two people would both rather be with each other than with whom they got. |
| proposer-optimal | The initiator lands the best partner the situation allows; the waiter lands the worst — same feelings, opposite luck. |
| a blocking pair | Two people quietly better matched to each other than to their partners — the affair a fragile arrangement invites. |
| flipping the proposing side | The road not taken: who reached out first, and the different life the other choice would have stably produced. |
The honest model
Each side is a handful of people, and every person holds a complete strict ranking of the other side, drawn at random when you make a new market — the lists you see printed are the real ones the algorithm reads. Deferred acceptance runs exactly as described: a free proposer asks the next name down their list; a receiver compares any new suitor against whom they currently hold and keeps the better; the loser is freed to ask again. The animation just replays that sequence of proposals one at a time.
Every number is measured from the resulting matching, not asserted. Avg partner rank averages each person's position of their assigned partner on their own list; proposer advantage is the receivers' average rank minus the proposers'. Blocking pairs is an honest exhaustive check — every left–right pair not matched together is tested to see if both would prefer each other — and it reads zero because the theorem holds, not because we forced it. Flip the proposing side and the same code re-runs; the tilt you see is computed, not staged.
Where the metaphor tears
The whole model assumes each person walks in with a complete, stable ranking of everyone else. Human preference is nothing like that: it is partial, contradictory, and — worst for the metaphor — made by the courtship itself. You often learn who you wanted by being pursued by them. When the rankings form during the process, "the best attainable partner" stops being well defined.
The tidy split into a side that asks and a side that holds is a fiction in real courtship, where both people signal, hedge, advance and retreat, and the roles swap by the hour. The theorem's sharp edge comes precisely from that clean asymmetry; blur it and the advantage smears into something real but unmeasurable — a lean, not a law.
Proposer-optimal only means best among stable matchings — and in a market where everyone's first choices collide, even the best stable option can be a distant one. The algorithm guarantees nobody has a partner-in-crime to run off with. It guarantees nothing about whether the person you're holding is anyone you'd have chosen with a free hand.