System OneENGINE

Pick a color palette from a design brief

Code checks contrast first. Jev compares the brief with the palette descriptions and can say that none fits.

  • Reusable recipe with code checks
  • Recipe palette-match with sysone_run
  • Recorded examples here. Live runs through your account or engine.

Recorded example: Night reading app. The interactive version loads with JavaScript.

Evidence and options

A reading app for long articles at night. Low glare and calm, with one restrained accent color.
  • Harbor: Deep navy surfaces, pale gray text and one sea-glass green accent. Low glare and quiet, made for long sessions at night. Passes AA contrast in code.
  • Studio Paper: Warm paper white with charcoal text and a muted terracotta accent. Calm and editorial, like a printed magazine. Passes AA contrast in code.
  • Arcade: Near-black violet with neon magenta and electric yellow highlights. Loud, playful and high energy, made for games and events. Passes AA contrast in code.
  • Orchard: Soft cream with forest green headings and an apple red call to action. Friendly and outdoorsy, made for community and family sites. Passes AA contrast in code.
  • Mist: Pale fog gray with light slate text and a powder blue accent. Airy, minimal and soft. Excluded by code: Body text on background 2.64:1, needs 4.5:1; Secondary text on surface 1.99:1, needs 4.5:1; Button label on accent 1.85:1, needs 4.5:1; Accent against background 1.64:1, needs 3:1.

Recorded Jev answer, September 24, 2026

  • Choice palette: harbor, 99%

Jev chose Harbor for this brief.

Apply Harbor for review

The pick meets your 80% rule and code already confirmed contrast. A person still reviews the rendered screen.

Measured when recorded: 292 ms in the engine, 1 model call, 524 input tokens. A recorded answer, not a live run.

Exact typed request for this example
{
  "state": "Brief: A reading app for long articles at night. Low glare and calm, with one restrained accent color.",
  "questions": {
    "palette": {
      "type": "choice",
      "instructions": "Choose the supplied palette that best fits the brief. Use review if none fits. Use review when no supplied candidate is suitable or necessary facts are missing.",
      "criteria": {
        "harbor": "Deep navy surfaces, pale gray text and one sea-glass green accent. Low glare and quiet, made for long sessions at night.",
        "paper": "Warm paper white with charcoal text and a muted terracotta accent. Calm and editorial, like a printed magazine.",
        "arcade": "Near-black violet with neon magenta and electric yellow highlights. Loud, playful and high energy, made for games and events.",
        "orchard": "Soft cream with forest green headings and an apple red call to action. Friendly and outdoorsy, made for community and family sites.",
        "review": "No supplied candidate fits, required evidence is missing, or the choice is ambiguous."
      }
    }
  }
}

One Choice question over the palettes that pass code checks, plus a review option for no fit.

How this experiment works

  1. Code Compute WCAG contrast for text, secondary text, button labels and the accent.
  2. Code Drop palettes that miss the contrast level or a required brand color.
  3. Code Render the chosen palette for a person to review.
  4. Jev One Choice question over the palettes that pass code checks, plus a review option for no fit.
  5. You Write the brief and describe each palette in words.
  6. You Choose the probability at which a pick is applied without review.
  7. You Treat review as a request for a new palette or a clearer brief.

Limits

  • Jev reads text. It never sees the rendered colors or the preview.
  • Contrast and brand colors are exact rules, so code checks them.
  • A probability is a model output, not measured design quality.
  • Recorded examples are real Jev answers replayed only for an identical request. An edit that changes the request needs a live run.

All experiments and community demos