A design system's real test isn't the launch — it's whether the second project that adopts it changes anything. Three months on, here's the honest tally for the Obsidian tokens.

Reused without changes

  • The four-step background scale (primary/surface/elevated/card).
  • Radius tokens — sm through full covered every case that came up, no fifth size requested.
  • The easing curves. Nobody ever asks for a new easing curve; they ask for the existing ones to be applied more consistently.

Quietly rewritten

The category/discipline colors shipped as five fixed hues. The second project needed seven categories, and rather than extend the token set, the team just picked two existing hues and doubled them up with a border-style variant. Not what I'd designed for, but not wrong either — a token system surviving contact with a use case I didn't anticipate is a decent outcome.

What I'd cut starting over

The glow/shadow tier had three levels (card, accent, glow-ambient) and only two ever got used independently — the third was always composed from the other two anyway. One fewer token, one fewer decision at the point of use.

The best sign a token system is working is how boring the retro is. This one was boring. That's the goal.