Design Language Audit — July 2026

Status: Superseded

This document is superseded as the counting mechanism. The drift script (scripts/design-drift.mjs) is now the source of truth. Where the script and this document disagree, the script is right. Every disagreement was in the direction this document's own lede predicted: understated.

This document remains in the record as the baseline measurement — the state of the product before Phases 0 and 1 landed. It is the "before" column in the verification table in the Phase 1 handoff.

The July 2026 baseline

Measured by hand against the full codebase on 2026-07-16, before any token work began.

What was measured

• 74 stylesheets across 16 remotes plus the shell

• 432 font-size declarations (35 distinct sizes)

• 158 button rule-sets across 13 recipes

• 34 badge treatments

• 6 spinner implementations with 4 different @keyframes names

• 30 box-shadow declarations (47% hardcoded, none mode-reactive)

• 25 z-index values spanning 0→200 with no scheme

• 99 unnamespaced selectors leaking across remote boundaries

• 33 dead fallback colours (fallbacks for tokens defined since May)

• 43 --color-bg declarations painting nothing on the live deploy

• 10 outline: none sites

• 123 native title= attributes (7 accessible)

• 11 product-wide accessibility defects

Key findings that shaped Phases 0 and 1

1. The ramps lied. --color__graphite-700 was darker than -900 on three of six greyscale ramps. Step numbers did not track luminance. This became gate A19, resolved in Phase 1.

2. Five light-mode contrast failures. --color-text-muted at 4.37–4.49:1 fell below 4.5:1 against two surfaces. This was gate R11, resolved in Phase 1.

3. Four phantom warn dialects. 22 declarations had independently invented --color-warning-*, --color-warn, and other variants for one concept.

4. Three competing focus treatments. Zero custom :focus-visible rings across the product.

5. Federation was a documentation convention. F1 forbade members from declaring tokens, but every member imported theme.css independently — the same race condition at runtime.

What the audit missed

• The Tier 1 consumption violations (F1a) — three files reading var(--font__mono) directly. The audit grepped --color__ only.

• The two script bugs (CRLF and argv indexing) that would later produce confidently wrong output.

• The CRLF frontmatter parse failure that silently emptied the member registry.

These are recorded in the Phase 1 handoff (§4) — they were found by the drift script that this audit's findings made necessary.