
The mark that tells a user a machine wrote this — monogram construction, the one gradient in the system, behaviour on every ground and the rules that keep it honest.
Our products increasingly fill fields on the user's behalf. A brand column pre-selected, a supplier value matched to the catalogue, a product type inferred from a description. The user is then asked to confirm — and confirmation is only meaningful if they can see, without reading anything, which values came from a machine and which came from the file or from their own hands.
Chapter 02 establishes that azure is the only colour that moves: it marks what is alive and operating. The AI signature extends that principle one step further. Azure says this is active. The signature says this was decided for you, and you can change it.
It is deliberately not a logo. It never signs a page, a document or a product. It attaches to a single value, and it disappears the moment a human overrides that value.
E — the letterform is lifted from the EurekaGO wordmark, not set in Archivo. It is the same vector used in Chapter 01, so the signature inherits the logo's construction rather than imitating it.
I — the official pictogram, which already reads as an exclamation mark, stands in as the second letter. Eureka Intelligence — EI.
Balance — the pictogram runs 60 units against the E's 34, roughly 1.8× taller, and the E is scaled down so its slab strokes thin out to match the pictogram's weight. Same letterform as the logo, matched optical mass.
Tile — a square with a 31% corner radius. Not a circle: two letters side by side need the horizontal room, and the squircle reads as a token rather than as a button.
Construction — drawn as a CSS mask over a gradient fill. No icon font, no image file, one element, and it inherits size from its context.
16 px is the floor. Below it the two letters merge and the mark becomes a blue dot — at that point use a text label instead. There is no upper bound, but above 32 px the signature starts competing with the logo and should be questioned.
The playbook is flat by conviction. This is the single exception, and it is what makes the signature unmistakable at 16 px: a radial fill, light at the core and deep at the rim, which keeps the tile reading as one object where a flat colour would dissolve into whatever sits behind it.
Because it is the only gradient, it needs no explanation in the interface. Anything that shades is the AI. Everything else in EurekaGO stays flat.
#233192 enters the
palette as --ai-deep, and the light stop is the brand azure already defined in Chapter 02 —
so the signature adds exactly one colour to the system. KOSMOS, the first product to carry it, has been
re-pointed at the canonical version; anything built from here inherits it. The rejected pair is kept
above as evidence, not as an option.
A linear gradient puts one pale corner and one dark corner on a 16 px tile: half the
monogram loses contrast against the white knockout. The radial fill keeps a dark rim all the way round,
so the two letters hold at every size. The mid stop #1268B4 exists only to stop the blend
from passing through a muddy grey.
One tile everywhere. The gradient never changes and the mark is never redrawn for a context — on dark or saturated grounds it gains a hairline white keyline, and nothing else.
Rejected alternatives, for the record: knocking the glyph out with no tile breaks the object; a translucent white tile loses glyph contrast; a white tile with an inverted glyph survives only over photography, where a keyline is not enough separation.
Four canonical placements, taken from the KOSMOS import wizard. Together they cover almost every case a EurekaGO interface will meet.
| Token | Value | Use |
|---|---|---|
--azure | #00AEEF | Light stop. Already in the palette, Chapter 02. |
--ai-deep | #233192 | Dark stop. New — the only addition this chapter asks for. |
--ai-grad | radial, 34% 26% | The tile. Reserved for the signature. |
--ai-bar | linear, 90° | Evidence bars only — a radial fill on a 6 px bar is meaningless. |
--ai-wash | #E8F7FD | Background of AI-owned suggestion chips. |
--ai-ink | #0E5E8C | Text on --ai-wash. Passes AA at 14 px. |
/* the whole component — no icon font, no asset, one element */
:root {
--azure: #00AEEF;
--ai-deep: #233192;
--ai-grad: radial-gradient(circle at 34% 26%, var(--azure) 0%, #1268B4 46%, var(--ai-deep) 100%);
--ai-glyph: url("data:image/svg+xml,…"); /* E of the wordmark + pictogram */
}
.ai-mark {
position: relative; display: inline-block;
width: 20px; height: 20px; border-radius: 31%;
background: var(--ai-grad);
}
.ai-mark::after {
content: ""; position: absolute; inset: 0; background: #fff;
mask: var(--ai-glyph) center / auto 60% no-repeat;
}
.ai-mark.sm { width: 16px; height: 16px; border-radius: 30%; }
.ai-mark.sm::after { mask-size: auto 66%; }
.ai-mark.lg { width: 26px; height: 26px; border-radius: 32%; }
.ai-mark.ring { box-shadow: 0 0 0 1.25px rgba(255,255,255,.92); } /* dark grounds */
The full data URI is in the reference implementation shipped with
the KOSMOS import wizard. Markup is one line:
<i class="ai-mark sm" role="img" aria-label="Suggested by Eureka Intelligence"></i>
The signature will spread faster than any other element in this playbook, because every product is adding machine-written values. Three things keep it from drifting.
| Question | Answer |
|---|---|
| Who may add a new placement? | Anyone, provided the value is genuinely machine-authored and editable. No approval needed — the rule in section A is the test. |
| Who may change the mark? | Brand only. Geometry, gradient and radius are fixed; a product that needs a variant needs a conversation, not a local override. |
| What about a different AI name per vertical? | The name is centralised in one constant in the reference implementation. Verticals inherit Eureka Intelligence; the mark does not change even if the name does. |
| Does the mark ever appear outside product UI? | No. Not on decks, not on the site, not in social. It is an interface signal, and it means nothing where there is no value to attach it to. |