Why this works

We built PartyUp on the published science of how people actually bond — and we'd rather show you the evidence than ask you to trust a vibe. The short version:

Matching algorithms can't predict chemistry. At all. In a landmark study, machine learning applied to 100+ traits and stated preferences predicted essentially 0% of who would actually want whom after meeting (Joel, Eastwick & Finkel, Psychological Science, 2017). Every "compatibility score" you've ever been sold is selling past that result.

What does work is boring and beautiful: repetition. Roughly 50 shared hours makes a casual friend, ~90 a friend, 200+ a close one — and leisure hours count more than work hours (Hall, Journal of Social and Personal Relationships, 2018). People randomly seated next to each other in one class session were measurably closer a year later (Back et al., Psych Science, 2008).

Novelty is an accelerant. Couples assigned to novel, mildly challenging activities together showed measurably higher relationship quality than those doing pleasant-but-mundane things (Aron et al., JPSP, 2000) — which is why our seasons escalate toward a finale instead of repeating a coffee meetup.

Swiping selects on exactly the signal that stops mattering. Strangers agree on who's attractive; that consensus collapses as people actually get to know each other (Eastwick & Hunt, JPSP, 2014). And the swipe interface itself trains rejection — acceptance drops ~27% within a single session (Pronk & Denissen, 2020). Meanwhile only about 1 in 10 partnered Americans met their partner on an app (Pew Research, 2023), and roughly 4 in 5 recent app users report burnout.

So we don't match people to people. We match people to tables. Same crew, same place, a season at a time, with a little adventure built in. The science builds the room. What happens at the table is yours.

Want the full bibliography — every study, every source, including the ones that cut against us? Email hello@joinpartyup.com and we'll send the research log, unedited.