Slate is the product. Color is the data.
Chrome is neutral. When color appears it marks a band, a status, or an evidence state, or it is the one brand accent, which is brand and never state. Semantic color is always paired with a label; never the only indicator.
Slate rail — canvas and raw fills
Slate 50 → 900Text and border tiers
Neutral oklch, not slatePrimary — near-black
Actions, focus anchorBrand accent — Editorial Plum
--eg-accent · brand, never stateThe accent belongs to the brand and marketing surfaces (apps/www, apps/docs/brand) — eyebrow dots, editorial links, article rules, OG cards, illustration highlights. Authenticated product apps use foreground and muted tokens for links and nav, not plum. Never on band pills, primary buttons, or inside the mark; the focus ring stays neutral. Utilities: text-brand, bg-brand-soft, border-brand-border.
Match-band aliases
--eg-band-* · BandBadge renders bg · fg · borderBand token reference
Full --eg-band-* familyStatus colors
--eg-status-* aliases · use these, not raw railsEvidence states
Dashed shells on criterion evidenceAbandoned directions
Do not reintroduce as brand or chromeOne family. A deliberate scale. Tabular numbers everywhere that counts.
Geist Sans for UI, Geist Mono for tabular and code-ish content. Product screens use RouteShell titles and review headers from the H1 and body rows below, not the marketing hero clamp (reserved for apps/www).
The scale
Geist SansEyebrow
.eg-eyebrow11px, semibold, uppercase, 0.15em tracking. Default color is fg-3; override with --eg-eyebrow-color for the rare dark variant, as shown above.
Tabular numbers
Stats, progress, countsCorners scale by role. Shadows stay light.
Shape is quiet; the point is content, not chrome. Resting surfaces carry a faint shadow, overlays lift further, and nothing glows.
Radii
--eg-radius familyShadows
--eg-shadow-* mapped in theme.cssA 4-pixel grid. One shared content column.
Card padding is p-6 (24px). The RouteShell stack is gap-6. Canvas padding is px-4, px-6 from lg, and py-6. Page mode keeps RouteShell at its default max-w-6xl; evidence review uses the wide max-w-7xl column. Everything else falls out of the 4-pixel grid.
Spacing scale
Tailwind-alignedTwo surface modes
Pages vs. workspacesDashboard, settings, credits, create flows, and setup wizards. RouteShell default width, max-w-6xl, on a px-4 lg:px-6 py-6 canvas. Card-based, page-level scrolling. Users pass through quickly.
Candidate review. RouteShell width="wide" (max-w-7xl): a 380px candidate list beside a sticky detail panel. The page scrolls as one. Users spend time here.
Lucide. Default stroke. Size follows the slot.
One icon set, stroke-based, at Lucide's default 2px stroke. Always paired with a label or used in a known slot. Sidebar rows use 18px, admin rows 16px, and dense review controls 16px.
Sidebar navigation
SidebarNavLink · 18 pxAdmin navigation
SidebarNavLink dim · 16 pxReview and chrome
16 px in dense controls · Menu 20 pxDurations by job. One named keyframe.
Consumer transitions use the durations below. Rotating UI uses the named keyframe.
Transition durations
Hover, panels, content swapsNamed keyframes
@keyframes shipped from tokens.cssReduced motion
The contract on `prefers-reduced-motion: reduce`- AnimatedGridPattern has no breakpoint of its own. Call sites hide it below their breakpoint (www: lg). It does not yet respond to `prefers-reduced-motion`.
- AutoCycleProgressBar and AutoCyclePauseToggle render nothing under reduced motion; the cycle does not advance.
- No fade-in on page load. Use `animate-in fade-in-0` only on content swaps the user triggered.
- No bounces, no springs, no chromatic ring transitions.
Visual rules and common mistakes.
Visual rules come first. They match the table at the top of PATTERNS.md. The tiles below pair the wrong pattern with the right one.
Visual rules
PATTERNS.md · selected row · quotes · tokensSelected list row
Wrong: a colored left bar, a left bar on a card, or a left bar on the active sidebar item.
Right: the neutral slate left bar only on the selected row in a list (`ReviewQueueItem`, Shortlist candidate list). Sidebar active stays font-medium and foreground.
Quoted evidence
Wrong: a left rule on the outer card, highlight bands pretending to be quote chrome, or a second quote style in the same flow.
Right: `EvidenceQuote`: bordered figure, blockquote with `border-l-2 border-foreground/15` inside the shell.
Surface fill
Wrong: a tinted background and a colored border on the same box.
Right: a background or a border, not both in competing colors.
Semantic color
Wrong: a green dot or amber strip with no label.
Right: color plus a label or icon (`Badge`, `BandBadge`, status dots).
Tokens
Wrong: text-[11px] or raw emerald-100 in new code.
Right: text-2xs, bg-verdict-*, and mapped shadows (shadow-xs, shadow-sm).
List elevation
Wrong: heavy shadows on dense list cards.
Right: shadow-sm at most on list surfaces.
Shipping
Wrong: a one-off composition because the screen is urgent.
Right: add the pattern to PATTERNS.md and the showcase, then import it.
Color on data only
Wrong: colored nav, tinted ordinary cards, colored borders, or text-brand in a product app.
Right: slate and semantic tokens for chrome. One exception: a featured upsell card, such as the recommended credit pack, may carry a sky tint.
Bands, not scores
Wrong: 87% on the review list, the candidate header, or a screening export.
Right: Suggested match on the list and header. The numeric breakdown sits in the analysis audit sheet, opened from the Criteria heading.
Active state: weight, not pill
Wrong: a pill background or a rounded highlight on the active sidebar item.
Right: font-medium and text-foreground on the sidebar. List selection uses the left bar rule above, not the sidebar pattern.
Empty states earn their keep
Wrong: a mascot, “Oops”, or an exclamation mark.
Right: two short sentences: what is here, and what to do next.