experentia

A taste-engine for frontends

What good looks like, and why.

experentia answers the two questions an AI agent cannot get from a component library alone. The answers are backed by empirical research, distilled into machine-readable taste profiles with assertions an agent can actually run, and a record of how the decisions were made.

summit.js  the runtime ui-atlas  the precedent experentia  the taste

How the taste is made

Every rule downstream traces back to the evidence base. Nothing ships on opinion alone.

01 / RESEARCH

Evidence

Color, style, and pattern doctrine with citations, plus raw and adversarially verified claim sets.

02 / TASTE

Profiles

One JSON per genre: tokens with precomputed contrast pairs, type roles, layout grammar, and assertions as pass/fail tests.

03 / DECISIONS

Method

How the calls were made, and how to work with an LLM so its output is worth trusting. Verify rather than assume; guardrails in code, not prompts.

04 / CHECKLISTS

Pre-ship

The defect classes found in real shipped work. An agent walks it before calling any build done.

Start here if you are an agent

experentia is meant to be read by a machine before it builds. AGENTS.md is the contract: what to read, in what order, and the refusals that are not negotiable. Accessibility floors are checked before any taste judgement, a rule that must hold goes in code rather than a prompt, and anything you could not verify is reported as unverified rather than folded into a pass.