A working index, 2017–2024.

The 2023 edition of the Ridgeline spec assumed consumers would tolerate a single, opinionated theme. In practice, six adopting design systems asked for a cleaner separation between structure and skin — a way to re-theme components without forking their accessible behavior. We rewrote the entire theming layer over six months, shipped a 140-page public handbook alongside, and migrated the four largest adopters in the same quarter the spec was published.

The 2.0 release was downloaded by 2,400+ teams within its first year. It remains the only production-grade component spec written for both TypeScript and Figma Variables from a single source of truth.

The Halden team came to us with a working model and a working prototype, and a working sense that the two could not be shown to a partner without the partner laughing at the interface. We spent the first month in their offices observing partners drafting, then rebuilt the drafting surface around a single, persistent document object the model could write into alongside the human. The previous product had been three separate panes; the shipped version is one.

Time-to-first-draft for new partners dropped from a week to under an hour. Halden closed their next round four months after launch.

A non-profit climate research arm with a sprawling set of indicator datasets, none of them public, and a board that wanted the work to be visible by the next grant cycle. We built the publication pipeline, the dashboard, the print companion, and the editorial workflow that lets their researchers publish without an engineer in the loop.

The dashboard has been cited in two subsequent IPCC working papers. The print companion was the more demanding of the two.

Reading a 1623 First Folio on a five-inch screen is a different problem than reading a modern paperback. We worked with the Folger's curators for nine months to design transcription-aware typography, a palimpsest viewer, and a citation system that survives the leap from physical page to mobile scroll. The app ships with seventeen digitized manuscripts and the institution's first-ever public annotation layer.

It was, against our expectations, the engagement that taught us the most about accessibility in the year we learned anything about accessibility.

A two-person founding team with a working radar aggregation layer, and no interface. We designed the visual language, the onboarding flow, the alert model, and the iOS / Android shells in eleven months. The product has no advertising SDK, no third-party analytics, and a thirty-page public privacy document that names what is not collected.

Pluvi passed 200,000 paid subscribers in its first quarter and has not added a single tracking pixel since.

Sintra's clinical product had grown across three acquisitions and was held together by twelve different shades of blue. We consolidated the system to a single token graph, rebuilt forty-two components to spec, and trained their internal team to maintain the system without us. The work shipped in eight months and now spans every Sintra surface including the patient portal.

The harder half of the engagement was convincing four engineering leads, in four time zones, that fewer components was the right answer.

A CLI and a reference web app for tracking how a column in a warehouse got to be the way it is. The team had built a working parser in Rust; we worked on the contract — the public command surface, the error messages, the docs site — for fourteen weeks. The CLI shipped with a 90-page reference manual we wrote in parallel with the engineering work.

The reference manual, not the CLI, was the part that took the longest.

A long-form literary quarterly, independent, with a thirty-year archive on a CMS no one at the journal could log into. We designed a new wordmark, a new typographic system, a reading-first archive interface, and a contributor portal that an editor-in-chief could actually use. The archive went online in eight months and now serves roughly one-third of the journal's traffic.

The identity was the easier half. The CMS replacement took twice as long as we had estimated and was worth every additional week.

A 300-person construction firm whose foremen were tracking daily site status in a group text. We replaced the text with an offline-first mobile tool, a dispatcher view, and a single shared ledger. The tool has been in continuous production use for two and a half years and was extended by Quarry's own team for the eighteen months following launch.

The hardest constraint was that the field had no signal sixty percent of the time. Everything had to reconcile later.

An open data portal for a regional transit network — schedules, service alerts, accessibility data, ridership, all in one place. We designed and shipped the portal in seven months and trained a four-person team to maintain it. It has been continuously published since, with no further studio involvement.

It is the engagement we are still most often asked about, and the one we are least able to talk about in public.

The first edition of the Ridgeline spec, written across twelve months with three contributing teams. It was the first public, production-grade component specification authored in TypeScript and Figma Variables from a single source. Six major design systems adopted it within its first year; the 2.0 edition in 2024 was written because the spec needed to outgrow its origins.

Ridgeline remains the studio's most-cited piece of work and the only one we did not invoice for.

A point-of-sale and inventory system designed for independent bookstores, by an independent bookstore. The product is still in active use across roughly eighty stores and has been maintained without studio involvement since 2020. It is, by some distance, the engagement whose continued existence in the wild we are most quietly pleased about.

It also taught us that "small" software is not the same as "easy" software.

An internal research project: what does a design tool look like when the canvas is a long-running agent rather than a static artboard? We built a working prototype over six months, used it to ship three subsequent engagements, and then open-sourced the agent layer. It is the work that, in retrospect, prepared us for every AI-native engagement that has followed.

It was not, in 2018, something we could talk about publicly without sounding unserious.