A human example
A creator asks for a calm, readable workspace with one restrained accent.
What the caller supplies
The agent supplies three approved palettes with names, hex values and plain-language descriptions.
What happens next
Jev returns a palette ID or review. The app renders a preview and checks contrast in code.
Illustrative example, not a recorded result.
Potential value: medium
Useful when a person already has approved designs and wants a relevant starting point. It does not create a palette or judge an unseen image.
Evidence confidence: low
A builder documents design and palette selection, but no blinded preference comparison is available.
The rating describes support for this claim. It is separate from Jev's returned probability. How we assign ratings.
Evidence, including disagreement
- Choose a card design from a page description. anecdotal. A concrete design-selection integration exists; preference quality is unmeasured.
The next test
40 original briefs with three to eight approved palettes, including conflicting preferences and no suitable option.
Compare against
- Fixed default
- Keyword matching
- Direct agent selection
Measure
- Blind human preference
- No-match recognition
- Total time and calls
- Deterministic contrast failures
Decision after the test
Keep as an optional recommendation only if blind reviewers prefer it to the default without increasing contrast failures.
The report will retain inputs, question versions, every attempt and failure examples. We will update the confidence rating after reviewing the result.
Use a related Engine recipe
Recipes are implementation starting points. Their presence does not mean the protocol above has passed.