STATION ·07
WHAT THIS TESTS
Twelve siblings, one shared class. Edits should propose a sibling group; the patch should set `siblingGroup`, not `null`.
An engineer-led product team tries to ship a marketing page on its own and learns three uncomfortable things.
Screenshots get stale, Figma comments get archived. A structured diff is the only review surface that survives the merge.
Before you migrate the codebase, count the number of edits that actually need props. The answer is usually small.
How the picker decides whether two elements are siblings, cousins, or just dressed alike for the photo.
It still has IE6 hacks. It still works. We asked it for advice on cascade order and it told us to read the spec.
When grep stops working, when refactors stall, and when the design system slowly becomes a synonym for `bg-zinc-900`.
Every `!important` you add is a future ticket. Every nested selector is a quiet promise to your future self.
We removed the staging gate for a quarter to see what would happen. What happened was mostly fine, and that was instructive.
Spoiler: we mostly do not. The handoff is replaced by a shared canvas and a willingness to fix the other person's mistakes.
A short essay on why the picker matters more than the editor, and why we spent a month on a crosshair.
The good news: most pages are mostly accessible. The bad news: the gaps cluster in predictable places.
Container queries, subgrid, anchor positioning. We promise to try harder. We have promised this before.