Skip to content

Confidence scores

Updated 2026-05-19

Every component page shows a row of confidence chips near the top:

Confidence anatomy high variants medium motion medium a11y high content low

These tell you, at a glance, how well-curated each facet of a component’s documentation is. A chip is NOT a quality score for the component itself — it’s a score for the documentation about the component.

Reading the chips

The five facets

FieldWhat it covers
anatomyNamed slots, layout regions, and structural parts (header, body, leading-icon, …). Drives the Anatomy section + design overlays.
variantsThe full set of variant axes (State, Size, Theme, …) plus the values along each axis. Drives the Variants matrix.
motionInteraction/animation patterns the component participates in (hover, press, expand, etc.). Linked from the Motion section.
a11yWCAG 2.2 AA requirements and the patterns the component should follow. Linked from the Accessibility section.
contentUI copy guidance — labels, helper text, error states, tone. Component-specific (not category-defaulted).

The three levels

LevelMeaningWhen you see it
high Curated + complete. Reviewed by the design team; data fully filled in. Treat as authoritative.Stable components with full guideline coverage.
medium Curated but partial. Reviewed but at least one expected detail is missing. Useful, but verify edge cases against Figma or with the design team.Components mid-curation; common during category-defaults rollout.
low Stub or unreviewed. Auto-generated from the Figma sync; no human curation yet. Trust the registry shape (variants, properties) but check Figma for everything else.Newly-synced components; stubs awaiting authoring.

Where the values come from

anatomy / variants / motion / a11y

These four are category-level defaults. Every component in a category inherits the category’s confidence baseline from components/dist/categories/<category-slug>-defaults.json in the knowledge repo. The design team sets the level per category as it curates that category’s recipe set.

Example (Action category — Action-defaults.json):

{
"confidence": {
"anatomy": "medium",
"variants": "high",
"motion": "high",
"a11y": "high"
}
}

Every Button, Link, Sticky footer page renders those four chips identically.

content

Synthesized per-component at page-generation time:

  • high when the component has a hand-curated guideline (components/src/guidelines/<slug>.json) AND that guideline includes a content_guidelines.sections block.
  • medium when the component has a hand-curated guideline but no content section.
  • low for auto-stub guidelines or missing guidelines.

When to trust a page

Roughly:

  • All chips high or medium → safe to use as-is for design + build work. Cross-reference Figma if you’re working on edge cases.
  • One or more chips low → the page’s structural data (variants, properties, Figma URL) is still correct, but the guidance may be incomplete. Treat the page as a starting point; talk to the design team for prescriptive answers.
  • All chips low → the component is freshly synced and the documentation is auto-stub. Trust only the registry-level facts (name, key, Figma node, variants). Everything else needs human input.

Improving a chip

Confidence isn’t a static rating — it bumps up as guidelines get curated:

  1. anatomy / variants / motion / a11y: bump by editing the category-defaults file in knowledge, then resync. Bumps lift every component in that category simultaneously.
  2. content: bump by curating the component’s guideline JSON to include a content_guidelines.sections block. Bumps just that one component.

The chip update reaches docs on the next vendor refresh (nightly cron, or manually via vendor-snapshot.yml).

  • Migrations — schema versions + parallel-change discipline; the contract the confidence data depends on
  • State — sync state + tag versions across the three repos