Build Output and Performance

facet build resolves the token graph, scans authoring inputs, and writes static artifacts. The output is zero-runtime CSS plus optional token exports for other targets.

facet build tokens/app.resolver.json --out dist
facet build tokens/app.resolver.json --out dist --format oklch --minify

Standard outputs

The guide lists these build artifacts:

File	Purpose
tokens.css	CSS variables per mode, emitted as rgb triplets or oklch
utilities.css	deduped atomic classes used by scanned code
tokens.dtcg.json	pure DTCG export
tokens.props.css	@property typing progressive enhancement
tokens.lightdark.css	light-dark() progressive enhancement
native/	React Native, Swift, Kotlin, and JS token files
tailwind.css	Tailwind v4 @theme preset
tokens.styledict.json	Style Dictionary v4 JSON
demo.html	simple generated demo
playground.html	mode and contrast explorer

Atom deduplication

Utilities, fx(), and recipes all lower to the same FIR. Identical declarations share one content-hashed atom, even if they came from different files or authoring surfaces.

{ bg: "action", fg: "on-action", radius: "control", px: "inset.lg", py: "inset.md",
  _hover: { bg: "action-hover" }, _focusVisible: { ring: true }, "@md": { px: "inset.lg" } }

The source example says this object intentionally duplicates the button utilities in usage.html, so it contributes zero new atoms for those shared declarations.

<button class="bg-action fg-on-action radius-control px-inset-lg py-inset-md hover:bg-action-hover focus-visible:ring @md:px-inset-lg">
  Primary action
</button>

Mode diff output

The resolver emits the all-default coordinate at :root, then emits only changed token variables under non-default selectors. For the checked-in graph, that means diffs for selectors including:

• .dark
• [data-brand="globex"]
• [data-brand="verdant"]
• [data-density="compact"]

Atoms reference CSS variables, so changing theme, brand, or density swaps variables without re-emitting utility classes.

Performance status

The design notes report that the owned JavaScript pipeline compiles 50,000 components in about 200 ms in the reference benchmark. The same notes call optimization an in-house future need, not a reason for a Rust or external compiler rewrite.

Treat those numbers as reference-implementation context, not a published service-level guarantee. Facet is still pre-release.