Arbor

Problem

Four squads, three button styles, two shades of the brand blue, and a lead engineer rebuilding the same date picker for the third time. Nobody had time to build a system, which is usually why one is needed.

Approach

I started from the screens that already shipped, not from a blank library: the twenty most used components, their real states, and a token set that mapped one to one onto the code. Every component page documents the decision, not just the anatomy. Dark mode was a token mapping, not a second library.

Outcome

New screens ship 2.4 times faster by the team's own measure, and the date picker has been built once.