FIELD EXPERIMENT / 01Synthetic people. Real algorithm.

THE FIRSTDATE IDEA, UNDER A MICROSCOPE

Stable ≠ equal.

Same people. Same preferences.
Change who proposes. Watch who benefits.

Who gets to propose?

Ready

Rankings read left → right, then down on mobile. Tap to move a choice to first place. Green = held partner; peach/rust = latest proposal participants/connection; crossed-out = excluded.

Proposal trace

    THE COUNTERFACTUAL

    Flip the proposer.
    Keep everything else.

    These are the final outcomes for the current rules. Lower rank is better. Average rank counts matched people only; unmatched people are reported separately.

    What the post gets right — and where the guarantee ends

    Confirmed by FirstDate

    The official GovTech pilot describes Gale–Shapley matching, one match per cycle, Singpass verification, 72 hours to respond and contact sharing only after mutual acceptance. It explicitly says compatibility on paper does not guarantee chemistry. Official FAQ ↗

    Stable, within a model

    No eligible pair mutually prefers each other to their assignments, given the ranked lists supplied. It does not mean everyone gets a first choice, that every real person was considered, or that relationships will work.

    The proposer advantage

    With strict two-sided preferences, deferred acceptance gives proposers their best stable partners and receivers their worst among stable outcomes, not necessarily bad partners. If the stable matching is unique, switching sides changes nothing. Nobel explanation ↗

    Claims not established here

    “First app”, deletion as its success metric, and outperforming commercial dating apps are not demonstrated by the sources. The 2012 prize went to Roth and Shapley, not to an algorithm. Kidney exchange uses related market-design methods, not simply this one-to-one routine.

    Simulation assumptions: two disjoint groups, four synthetic people on each side, strict rankings, fixed preferences, capacity one, no ties, no learned compatibility model. Groups are mathematical roles, not genders. Hidden-edge and cutoff controls are experiments, not FirstDate implementation claims. We do not know FirstDate’s score-to-rank mapping, proposer assignment, tie-breaking or exclusions.

    Original post ↗ · NRMP algorithm ↗ · 2012 prize ↗