Blog · July 5, 2026
How We Pick
One's Vibe now scores every published project against a written rubric to nominate Editor's Pick candidates. Today we're publishing how it works — not because the scores are public (they never will be), but because the rubric is a statement of values. If you make things, it doubles as a map: here is exactly what this gallery rewards.
Two gates, very different heights
Getting in is easy on purpose. The admission bar asks three questions — can a visitor use it in the browser right now, is it what it says, is it safe — and when in doubt, we publish. Publishing is not an award; it's a fact: this thing is live and tryable.
Featuring is the opposite. The Editor's Pick spotlight is the only quality signal on the site, so it has to stay expensive. Most projects in the gallery will never be picked, and that is working as intended — the Explore page, tech-stack filters, topics, and collections exist precisely so every project can be found by the people it was made for, without needing our blessing.
This philosophy has a name here: “One's vibe, another's poison.” Value is subjective — your poison may be someone's vibe, so the door stays open. Taste has standards — so the spotlight stays expensive. Disagreement isn't a failure of curation; it's the point of a gallery.
The six dimensions
30-second aha — We are a try-it-now gallery. The single most valuable thing a project can do is let a stranger feel something within thirty seconds — no signup, no configuration, no demo video standing in for the real thing. This is the heaviest weight because it is the promise of the whole site.
Craft & UX — Did someone care? Typography that was chosen, empty states that were designed, copy that was edited. Important nuance: rough is not disqualifying. A plain-looking game with a brilliant core loop loses points here and wins them back elsewhere. Craft is an axis, not a gate.
Unique & memorable — “Would I send this to a friend?” An idea we haven't seen, or a familiar one executed with an angle that sticks. The judge compares against everything already in the gallery — the hundredth resume optimizer starts from a lower ceiling than the first playable weird experiment.
Real value — Does it solve something for someone imaginable? Deliberately the lightest LLM-judged weight, because value is the most subjective thing here — and because fun is value: a genuinely fun game scores as well as a productivity tool.
Transparency — Scored by formula, not by opinion: claimed your maker page? Filled in a Build Receipt — team, time, tools? Linked a public repo that actually resolves? Each is worth real points. This is the incentive we most want to exist: makers who show their work should win ties against makers who don't.
Alive — Also pure data: your uptime streak (the “Nd up” badge on cards), recent commits, no failing checks. A project that stays up for weeks quietly outranks one that flickers. Our graveyard is proof we take this seriously.
What we deliberately do not score
- How much AI you used. There is no purity test in either direction — “built with AI” is context we display, never a badge we grade. A hand-written project and a fully AI-generated one compete on identical terms: what does a visitor experience?
- Follower counts, launch-day upvotes, hype. The discovery pipeline finds projects in “drop your project” threads regardless of whether the maker has 148 followers or 130,000 — the rubric never sees either number.
- Score visibility: individual scores stay editor-only, forever. A public number would become a ranking, a ranking would become the product, and the long tail — the whole point of curated abundance — would die. The score's only job is to surface 25 candidates for a human to consider.
Calibrated to be stingy, decided by humans
The judge is instructed to keep the gallery median around 4 out of 10; a 7 should be top-ten-percent rare. The candidate pool is the rolling top 25 never-picked projects — no absolute threshold, so the bar rises as the gallery gets better. Once picked, a project retires from the ranking into a re-pick bench: spotlight seats keep going to new faces.
And the final call is never automated. The AI nominates; a human picks, balances categories, favors the undiscovered, and rotates picks weekly. When the judge isn't confident — say, it can only see a static screenshot of an interactive app — it says so, and that uncertainty is displayed to the editor rather than silently guessed away. We learned this the honest way: our judge once rejected a working product because its dark-themed page loaded slower than our screenshot. It got a formal apology, a slower camera, and a rule: blank evidence means look again, not reject.
If you're a maker, here's your checklist
- Make the first 30 seconds effortless — the core action, reachable without an account, is worth more than any redesign.
- Claim your project (sign in with X) — it's the single biggest transparency jump.
- Fill in your Build Receipt: team size, build time, tools. Facts, not marketing.
- Link your public repo if you have one. We verify it resolves — a dead link counts as nothing.
- Keep the URL alive. The uptime streak does the talking.
The full pipeline — from a maker's post on X to the spotlight — runs on this rubric today. Browse what it surfaces on the homepage or start where everyone starts: submit your project. It takes one URL.
“One's vibe, another's poison.”
Comments
Sign in to join the discussion.