Case Study:
The quiet ritual of a graduating campus, in numbers.
For three days in May 2026, the graduating batch of NIT Hamirpur opened a small private window and admitted, anonymously, who they had noticed across four years of college. This report is the honest read on what happened: how many signed up, who they picked, who picked them back, what the platform cost to run, and what the data says about reach, participation, and the shape of demand at one Indian engineering campus.
The picture in one page.
The roster covers 938 graduating students. About 21.4% of them signed up. Together they sent 238 crushes, and 11 pairs turned out to be reciprocated. That is the entire experiment, and every number that follows is a sharper view of those four lines.
Gender, the obvious first question.
The campus itself skews male. The signup pool skews more male. And the crushes that flow inside the system skew more still. Three nested asymmetries, each one wider than the last. The data does not flatter, but it is the honest portrait of an engineering campus in India in 2026.
Roster vs. signups, gender split
Share of each gender that signed up
Crush flow, by gender pair
Avg sent vs. received, by gender
The roster is roughly 79.2% male and 20.8% female. Among signups, that tilts further to 76.1% male and 23.9% female. Conversion is sharper still: 24.615384615384617% of female students on the roster signed up, versus 20.592193808882907% of male students. A smaller pool of women is louder per head. Crush direction follows: 171 Male to Female versus 6 Female to Male crushes flowed across the gender divide. Within gender crushes were rarer (41 Male to Male, 20 Female to Female), but non zero, and that matters as a sign the platform was used as intended: as a quiet honesty channel, not just a dating layer.
Demand is rarely flat.
The same phenomenon visible on every social network shows up here in miniature: a small share of profiles attract most of the attention. That is not a value judgement on anyone. It is just how aggregated noticing works inside a closed group.
How many crushes a typical active user sent
How received crushes are distributed
Across the people who received at least one crush, the top decile of profiles attracted 24.8% of all crushes sent. The median active user sent a handful of picks, not the cap of twenty. The shape is not surprising, but the magnitude is interesting for a closed campus where everyone knows roughly everyone.
The mutual moment.
A platform that only counts likes is a leaderboard. Double Dart only reveals when both sides clicked. That distinction is the entire contract, and the mutual count is the only honest test of whether the contract pulled real signal.
11 mutual pairs.
16 people walked away
with a name on May 14.
Mutual pairs, by gender combination
The arithmetic of reciprocity
Of every 100 crushes sent into the system, roughly 9.2 came backas part of a mutual pair. That number is the honest answer to “does it work”. For a graduating batch where many will scatter to different cities within weeks, even a single confirmed name is product market fit. 11 pairs is 16 people who now have one.
Branch tells its own story.
The shape of signups by branch mostly tracks the shape of the roster itself. Where it diverges, two patterns appear: branches that punched above their weight on signup, and branches that pulled disproportionate crush attention per head regardless of their size. The first is about reach. The second is about desirability.
Top branches, by signups
Top branches, by crushes received
Desirability index
Crushes received per student in branch. Branches with under 5 students excluded.
42.4%of all crushes stayed within the sender’s own branch. That is meaningful: at a residential engineering campus, your branch is your village, and the data confirms most quiet noticing happens inside that village. Cross branch attention exists, but proximity wins.
Did people actually use it.
Signups are a soft metric. The harder one: did the people who signed in send at least one crush. The answer determines whether a platform is being browsed or actually used.
What it took to run.
The whole product ran for three days on Supabase plus Vercel, with no dedicated infra team and no paid distribution. The numbers below are pulled straight from the dashboards. They are useful as a sanity check on how lean a campus scale experiment can actually be.
Vercel: who actually visited.
The site itself was almost entirely a mobile experience. Total visitors and page views are an order of magnitude above signups, which is the expected funnel shape: many came to look, a smaller group actually verified and used the product.
Daily visitors (May 7 to 14)
Daily page views (May 7 to 14)
Device breakdown
Operating system breakdown
85% of all traffic was mobile, dominated by Android at 66% of visitors. The peak was May 11, when word of mouth on campus put roughly 330 visitors and 1.3K page views through the site in a single day. After the spike the funnel held: bounce rate normalised to around 40% across the remaining days, suggesting people who came back were intentionally returning, not cold browsing. The Supabase load was correspondingly modest at ~26,904 requests per day, almost all of which were authentication traffic (sign in flow plus token refresh). The compute footprint of the actual product logic was effectively a rounding error against the auth chatter.
The honest read.
What worked
A three day experiment on a single campus, with no paid marketing and only word of mouth inside the batch, produced 201 verified signups and 11 mutual pairs. For a closed audience of 938 students, that is a conversion shape most B2C products would happily ship.
The privacy contract held. No one ever saw a one sided crush. That is the only feature that mattered, and it shipped without incident.
What did not
Reach was capped by reality. Roughly 78.6% of the batch never opened the platform. The gender skew on signups is real and is the single biggest constraint on match rate: a denominator that does not show up cannot reciprocate.
Of the people who did sign up, a portion never sent a crush. Signup intent and action intent are not the same thing, and the gap is worth designing around in any v2.
A graduating batch of around 938 produced 11 mutual pairs in three days. That is 16 peoplewho, on the day before they scattered, were told by name that someone had been noticing them back. The product’s job was to make that exchange possible without anybody else ever finding out, and on that single metric, it shipped.
Method and limits.
Where the data came from
Three Supabase tables: the official roster of the 2026 graduating batch, the profiles created via @nith.ac.in Google sign in, and the crush log (anonymous sender to anonymous receiver pairs). All metrics in this report are aggregates. No row identifies any individual, including in the desirability and mutual sections.
What to be careful about
Gender is taken from the institutional roster and normalised into Male, Female, Other, and Unknown buckets. Branches with under 5 students are excluded from the desirability index to suppress small sample noise. Signups are counted by created_at on the profile row, not by Google auth log.