An explorable explanation

BELLA, the system behind this site

A design system is only as useful to an AI as it is readable by a machine. This is that argument, demonstrated on the system running the page you are reading.

6 min read

Design systemsAI-enabled designGovernance

01 · AI readiness

Can an AI build with your system?

The difference is whether your system is machine-readable.

Flip the switch and watch the same agent change its mind.

Off. The agent is guessing.

What the agent built

An off-brand travel booking card an ungoverned agent produced. It reads Book your trip, Fast and easy booking today, and BOOK NOW. Clashing magenta and lime colours, a dashed neon outline, cramped spacing, body text below the readable floor, and a call to action whose lime-on-magenta text fails contrast.

The gate's verdict

Refused by the gate

4 of 4 checks failing

  • Invented a colour that is not in the system
  • Hardcoded a body size below the readable floor
  • Contrast fails at the accessibility bar
  • Restyled the button instead of using the component

No contract, so it invents what it does not know.

A documented pattern, not a measurement of this site. Southleft on context-based design systems

02 · Author, then enforce

Authoring is human. Enforcement is deterministic.

Move the control and watch the value travel.

A human decides what the token should be; everything after that is measured in this tab as you move it, and none of it is a recording.

01 Source

A human decision.

The control writes the custom property the page already resolves from.

reading

Shipped value, no drift.

--color-ink-soft: reading

02 Manifest

Generated, never typed.

Shaped by the same dtcgToken() that builds /api/bella.json.

reading
"--color-ink-soft": {
  "$value": "reading",
  "$type": "color"
}

03 Gate

Measured, not judged by taste.

Real relative luminance maths, against the AAA bar of 7 to 1.

The system inspecting itself.

reading to 1

Measuring

audit:contrast is reading the running stylesheet.

Agents read this same artifact, /api/bella.json, alongside the route map at /llms.txt. audit:agents fails the build if either one disagrees with the live registry.

03 · Enforcement

A system that cannot refuse is a suggestion.

Seventeen checks run before anything ships.

17checks, every one of them able to stop a merge

audit:structure

One route tree per case, the 1240 container everywhere, no arbitrary pixel classes, nothing off palette.

Where it lives
scripts/audit-structure.mjs
How to run it
npm run audit:structure

04 · Self-assessment

Where the system honestly stands.

Not a scoreboard. Strength in one place, room in another.

Scored against zeroheight's six-axis model. The org-scale axes are early by design; the frontier axes run deep.

Clockwise from the topFoundationsDocsGovernanceAdoptionMeasurementAI Readiness
StagesV1 → Growing → Teenage → Healthy ProductModel: zeroheight Design System Maturity Model. Self-assessed by Elleta McDaniel, ctrl_alt_design.
  • Foundations

    Teenage3/4

    Three token tiers, a 4px grid, AAA-minded contrast, zero hard-coded values.

  • Documentation & Knowledge

    Growing2/4

    Every token carries machine-readable metadata, plus DESIGN.md, a bella.json rollup and an llms.txt map.

  • Governance & Team

    Growing2/4

    Governance as code: a 17-audit gate that fails the build on drift, run by a team of one.

  • Adoption

    V11/4

    Powers elleta.design today with CHIP next, and few consumers by design.

  • Measurement & Impact

    V11/4

    Deliberately no vanity metrics; the working system is the evidence.

  • AI Readiness

    Teenage3/4

    The newest test of a design system is the path an AI takes, and this one defaults agents into it.

05 · The claim

A design system is AI-ready when a machine can build with it correctly.

Every claim on this page is a link you can open.

The manifest below is generated from the same source this page renders from. Read it the way an agent would.

What this stands on

Thanks for reading

Next case

CHIPAI + Design Systems · 2026