How we hold the line on the gallery — the small catalogue, the long rejection log.

The catalogue and the rejection log, the discipline that decides which is which.

Every quarter we open a candidate list of about a dozen pages we are seriously considering for the gallery. About a third make it. The rest get a polite "not yet" and a note explaining what would have to change for the answer to flip. Over the past eighteen months that process produced the catalogue you see today and a rejection log roughly three to four times its size.

The rejection log is the more interesting document. The work that almost made it is where you can see the criteria most clearly — what we let in is half the answer, what we kept out is the other half. We are sharing some of the entries below, with the company names redacted because rejection is private and editorial.

The five criteria, in order of weight.

First: typographic discipline. The page must hold one or two type families and use them with the consistency of a small editorial constitution. Three families on a single landing surface is a soft no unless the third is a deliberate, named voice (a code monospace counts; a decorative third-display does not). Linear, Stripe, Resend pass without argument. Most enterprise SaaS pages fail on this criterion alone.

Second: palette role coherence. Eight to twelve colors, role-tagged, with brand-tier flags clustering inside the brand vocabulary rather than scattered across decorative chrome. We surface this in the JSON via the isBrandTier flag; the editorial test is whether the flagged colors actually feel like the brand when isolated.

Third: motion vocabulary. The page either has a coherent motion language — one to two easings, two to three durations, used consistently across hover and scroll — or it is staying out of the gallery. Page-flooding scroll animations that nobody asked for are an automatic disqualification.

Fourth: contrast pass rate. Every primary fg/bg pair has to pass WCAG 2.1 AA on normal text. We allow APCA Lc above seventy-five for body, above sixty for secondary; below that we ask why. "Brand designed it that way" is not an answer that gets the page in.

Fifth: structural readability. The component tree has to be sane. Pages built as a single hero illustration with no semantic landmarks underneath get rejected even when they look beautiful, because the agent will not be able to write a parity component from a div soup.

The honorable rejections, with the names removed.

• A wildly popular dashboard product. Beautiful page; fails criterion one (four type families across the marketing surface, including a serif used decoratively for one heading and never again).
• A YC-batch fintech with strong brand. Fails criterion four (primary CTA contrast at 3.8 ratio, AA fail on normal text, the entire palette is keyed to a brand purple that does not survive in dark mode).
• An AI tool we genuinely love using. Fails criterion three (six different easings on a single page, durations ranging from 80ms to 1.4s with no apparent system).
• A respected developer-tools brand. Fails criterion five (the marketing site is a single video element with text overlays — agent has nothing structural to read).
• Three pages that are technically excellent but sit on the same editorial axis as a curated item we already cover. We will not catalogue parity. The gallery is not a tier list.

What the log becomes, and the slow shape of the gallery.

The five-criteria pass rate is intentionally low. We expect the gallery to grow slowly over the next eighteen months, then stabilize. Mobbin will scale to thousands. We will not. The gallery exists to be a small, dense, editorially honest set — the size at which a single curator can hold every item in their head and tell you, without checking notes, why it is on the list.

If your work is on our candidate list, you will probably hear from us first. If your work was rejected, the note exists in our archive, and the next quarterly review is the next chance. Reasonable disagreement is welcome — we have changed our minds twice in eighteen months. Both times the page improved before we wrote it up.