You Go First

A speculative dating-app concept built around scarcity, transparency, and portable trust. Fifteen picks a week, in three tiers everyone can see.

Problem: Swiping makes attention free, so a like tells you nothing and a few profiles get most of them.
System: A matching protocol on AT Protocol identity and Glowrm’s trust layer. Everyone gets fifteen picks a week in three tiers, and the other person sees the tier.
Mechanics: Scores are asymmetric. What you want (ten chips across five traits) is scored against what they are (a behavioral quiz): 0.7 × alignment + 0.3 × temperature. Conduct reported to Glowrm sets how many picks you get.
Stack: React, TypeScript, Tailwind and shadcn/ui; Supabase; AT Protocol OAuth.
Result: A working prototype at heypbj.xyz, with a simulator that runs both side by side.

Premise & motivation

The dating-app industry is a $12B market with ~80% churn within ninety days. Every major product is a variation on one mechanic: infinite supply, zero-cost swiping, photo-first ranking. Engagement is revenue, so the platform earns more when you don’t find someone. Users spend ~90 minutes a day swiping; ~75% of matches never exchange a single message.

YGF’s thesis is that the problem is economic. If attention is infinite and free, no signal a user sends carries information. So YGF treats attention as a finite resource: every user gets the same fifteen picks every week, split across three visible tiers: five Firsts, five Seconds, and five Thirds. Picks expire Saturday at 11:59 PM, a new roster lands Sunday at 7 PM, and there is no way to pay for more, so a match that cost a slot carries more information than a free like.

The premise section on a deep ink background, reading It is a dating app, shaped like a draft — The premise, as it reads on the marketing site.

The mechanic is called Stratified Preference Allocation (SPA), and it diverges from conventional preference matching in three ways:

Classical stable matching (Gale-Shapley) assumes a closed market, full information, and unlimited proposals. Real dating markets violate every one of those assumptions.
App-style preference systems (Hinge, Tinder, Bumble) collapse the signal to a binary like that is free to send and free to ignore. With no opportunity cost, a like carries no information. Hidden behind a paywall, it carries negative information.
SPA makes the cost finite through the fifteen-slot weekly roster and visible through bilateral transparency: when you draft someone First, they see “You are one of her five Firsts this week.” There are no hidden likes or paywalled superlikes.

Three principles fall out of that:

Strategic scarcity. Finite slots turn swiping into resource allocation. You can’t First everyone, so you have to choose. A First tells the recipient far more than a like does.
Bilateral transparency. Recipients see exactly which tier you placed them in. Both sides take the same risk.
Portable trust & identity. Reputation and identity are decoupled from the platform (ATProto for identity plus a behavioral trust layer), so good behavior follows you across apps and there’s no lock-in.

How a week works

FIRST · ×5

The most expensive signal you have.

When you draft someone First, they see it. There are five, and you can’t buy a sixth.

SECOND · ×5

Strong interest, one tier below the top.

Seconds are visible to the people you spend them on, at the right cost.

THIRD · ×5

On the radar, and the cheapest pick.

Thirds are how you keep someone in mind without overcommitting.

Three tiers, fifteen slots, one week, laid out as three columns of five — The mechanic as the marketing site lays it out.

The cadence is a week, from the Sunday 7 PM drop to the Saturday 11:59 PM lock, and between those two moments you spend your fifteen or you don’t.

A weekly calendar. Sunday: the drop. Monday: first moves. Tuesday: first reveals. Wednesday: mutuals. Saturday: lock. — One roster for seven days before the reset.

Design

Most dating apps borrow from Spotify, Apple News, Dribbble, or Steam. YGF borrows from print.

Source	Borrowed pattern
Sunday New York Times	Editorial mastheads, pull quotes, eyebrow labels
Old almanacs & specimen books	Ordinal numbering, asymmetric grids, dignified hierarchy
Magazine cover design	Display italics at billboard scale, generous whitespace
Newspaper sports section	Mono-numeric stats, agate type, terse data presentation

Four design decisions carry the system:

Old Standard TT italic at display scale, with a crimson period signing every wordmark and signal. The dot is the brand mark, and the serif does most of the visual work.
A constrained chromatic palette: ink, pulp, cream, plus three signal colors (crimson = First, marigold = Second, cobalt = Third) and a field green reserved for mutual drafts. Each color has one job. Crimson appears only in First-pick moments.
Duotone portraiture. Every photo runs through a tier-mapped duotone. Untinted photos read as Tinder, and the tint carries the tier with the person through the product.
Editorial copy as system. Voice is documented in DO / DON’T cards in the brand bible alongside type and color specs.

The Roster Board: fifteen slots arranged as three rows of five, with tier-mapped duotone portraits in the First row — The Roster Board, symmetrical so you can count at a glance.

Color · from the brand bible.

Type specimen page: Instrument Serif, Bricolage Grotesque, JetBrains Mono — Type system · three families, each with one job.

Numerals & ordinals: three display classes of numerals — Three display classes of numerals.

Two more surfaces

The List

A fourth verb beneath Pass / Third / Second / First, with a dashed-border treatment that telegraphs private, no cost. The List is unlimited, and the person you parked never sees it; the pressure comes from the Saturday lock.

This is the verb for “I don’t know yet, I want to think about it,” which is the most honest response anyone has to most cards.

The Night Round

A 9 PM to midnight synchronous window, signaled by a pulsing crimson banner and a city-wide push notification. Inside, you can send up to ten multiple-choice questions to people you’ve considered but not yet picked. They tap one of four answers.

There is no chat anywhere in it. The synchronous window is borrowed from BeReal.

The live prototype: discover view, picking Theo A., 34 — Live prototype · the draft surface.

The live prototype with the Night Round banner active — Night Round open · +List verb appears, two people are asking you.

Mechanics & implementation

The matching engine is asymmetric: your score for someone differs from theirs for you.

Relay assessment: a behavioral quiz produces ten vertex scores (Reach↔Retreat, Voice↔Quiet, Spark↔Steady, Merge↔Orbit, Heat↔Cool), normalized into a single Temperature score on a 0 to 80°R scale (e.g. 64.7°R). The integer is public; the decimal is a private hash used only by the algorithm.
Plinko priorities: users allocate exactly ten chips across five traits (attractiveness, liveliness, curiosity, kindness, adaptability). This captures what you want, is locked permanently, and is never shown publicly.
Score: matchScore = 0.7 × plinkoAlignment + 0.3 × tempCompatibility, where alignment maps your wants onto their derived traits and temperature rewards similar operating styles.

Supporting systems

The Roster. Fifteen visible slots arranged as three rows of five (Firsts / Seconds / Thirds), plus the unlimited private List. The Roster locks Saturday at 11:59 PM. Sunday at 7 PM, a new one lands.
Stratified visibility. Under-the-hood tiering recalibrated every fourteen days based on match success, within ~100-mile geographic regions so a small-market user isn’t ranked against a metro. It never appears in the UI: there are no letter grades, no public ladder, and nothing on profiles.
Behavioral ranking. Ghosting, unmatch rate, and report rate degrade roster access; consistent quality interactions improve it. Ranking points decay over fifty-two weeks to prevent indefinite coasting.
Draft picks. Selective cross-tier visibility that lets the engine surface someone outside your usual slice when compatibility is unusually high.
Safety & trust. Reporting, response-rate tracking, and ghosting detection feed a portable reputation layer that survives if a user moves between SPA-compatible apps.

YGF also includes a research artifact: a simulator and writeup analyzing the SPA system: attention-inequality metrics (Gini ~0.42), match-quality convergence over twelve weeks, behavior under continuous market churn, local-stability analysis.

The system, on a phone

The system at work: the Roster screen and the player-card screen on iPhone frames — Four mobile surfaces.

App Flow: eight mobile screens covering one week, from Sunday Reset to Onboarding — Eight hi-fi screens covering one week of use, presented on a pan/zoom canvas.

Every profile is a numbered card. Front: portrait, name, ordinal. Back: a manual of me, three things they’d cook you.

Voice

The brand bible specifies voice the same way it specifies type and color.

How YGF compares with other apps

A two-column comparison table: YGF vs every other app — The comparison panel from the marketing site.

What it costs

The five artifacts

The product surfaces as five connected pieces, each shippable on its own.

Brand Bible. A thirteen-section specimen book covering wordmark, type, color, voice, portrait treatment, the Roster, profile cards, and DO / DON’T copy samples. Written like a printed manual.
App Screens. Eight hi-fi mobile screens covering one week of use, from the Sunday Reset to the Saturday Recap, presented on a pan/zoom canvas.
Interactive Prototype. The full matching mechanic, playable end-to-end. Draft Pass / List / Third / Second / First, watch scripted inbound reveals land, hit a mutual, open a thread, send a message, and trigger the Night Round to send multiple-choice questions with AI-suggested options.
Marketing Site. A nine-section editorial landing page with a working waitlist form.
Strategy Deck. Twelve slides arguing the product, from the market problem to GTM (Portland Vancouver, WA Seattle) to projected milestones.

I did all of it: protocol design, product strategy, UX, brand, copy, and the deck.

Stack

App: React 18 + TypeScript, Vite, Tailwind + shadcn/ui, React Query, React Router, deployed as a PWA.
Backend: Supabase (Postgres 15, Deno edge functions, Realtime, Storage) with row-level security.
Identity: ATProto OAuth. Profile data lives in the user’s own PDS.
Trust layer: Cross-app reputation, shared moderation, and SPA delegated to Glowrm, the trust infrastructure layer YGF was designed against.
Type: Old Standard TT, Bricolage Grotesque, JetBrains Mono.
Hosting: Vercel.

Status

YGF is a portfolio piece and the blueprint for a hypothetical funded build. The matching engine, prototype, brand bible, app screens, marketing site, and strategy deck all ship. Production infrastructure (Sentry, cron jobs, SLAs, mobile apps) sits in the roadmap.

Trust layer: YGF’s portable reputation, ghosting detection, response-rate tracking, and global ban behavior all run through Glowrm, and YGF is one of three reference implementations of that API.