Tasteroll (roll for taste, keep what sticks)

Tasteroll audits what exists, asks at most three questions, fixes what is broken, then rolls valid alternatives. Lock what works; re-roll what does not. Project constraints produce taste even when choices are random.

The pipeline

1. AUDIT     → scan what exists → mandatory findings (craft-only)
2. INTAKE    → infer from context/repo OR 3-question mini-interview
3. FIX       → resolve all findings (seed-independent, non-negotiable)
4. GENERATE  → fresh candidates for open dimensions (context-aware)
5. ROLL      → seeded pick between valid alternatives
6. LOCK      → keep what works, re-roll what doesn't, converge
7. GATE      → relaxed gate (readable + WCAG AA + self-contained + inert)

1. Audit (mandatory fixes)

Scan rendered surfaces, source CSS, repo files, and content. Resolve every non-negotiable craft finding before rolling:

• Contrast: body < 4.5:1 or UI < 3:1 on actual surfaces
• Spacing: no scale, arbitrary margins with no token system
• States: interactive elements missing hover/focus/active/disabled
• Tells: named AI slop patterns present (gradient, centered hero, 3 cards)
• Accessibility: missing alt, unlabeled controls, broken heading order
• Structure: missing main landmark, no skip link, no focus-visible

Flag findings outside craft scope (bad copy, content gaps, broken features) and hand them off; tasteroll does not fix them.

A single-word "preserve" overrides an intentional low-contrast or missing-state choice; otherwise resolve every finding.

For greenfield, scan brief/content for contradictions, missing scope, and vague direction.

2. Intake

Choose by available signal:

Infer (default): read the conversation and scan repo files (README, CSS, package.json, content). Skip questions when product and audience are clear.

Mini-interview (fallback): ask three questions when inference is thin:

1. What are you building? → unlocks structure, density, IA
2. Who is it for? → unlocks personality, accessibility posture, complexity
3. One word for how it should feel? → unlocks aesthetic, accent direction

Three answers cover the brief without the full 11-field design-system-interview.

3. Fix (deterministic)

Resolve every finding here. Separate seed-independent "resolved findings" from random "rolled choices" in the output.

4. Generate

For each open dimension (personality, aesthetic, type, color_mode, density, rhythm, signature, imagery, motion, accent), generate 2–5 candidate options from audit/intake context at roll time; no static list.

Candidates must satisfy compatibility rules in assets/design-rails.json: compact density/metronomic rhythm, airy/syncopated, brutalist/restrained motion, etc. No conflicts.

Respect locked dimensions from previous rolls.

5. Roll (seeded pick)

The PRNG (assets/tasteroll-engine.js) picks one candidate per open dimension. Same seed

• candidates → same roll, always; seeds make rolls reproducible/shareable.

If Chance (kyanitelabs/chance) is available via MCP, prefer it for multi-source entropy mixing and reproducible audit trails; the inline engine is the fallback.

6. Lock and re-roll

Mode:

• One-shot: roll once; take it or re-roll from scratch.
• Iterative: roll → review → lock working dimensions → re-roll unlocked only → converge. Each re-roll increments the seed and excludes shown directions.
• Shotgun: roll N seeds → compare side by side → pick one to develop.

Locked dimensions and resolved findings persist; only open dimensions vary.

7. Gate

Tasteroll output is a personal artifact/prototype, not shipped; use a lighter gate than tastecheck-pass:

• Readable: start with body ≥ 1rem, measure 58–75ch, and line-height ≥ 1.5; a brief may justify another measure with rendered evidence
• WCAG AA: body ≥ 4.5:1, UI ≥ 3:1 (verified on the rolled surfaces)
• Self-contained: no external requests in exports
• Inert: no scripts in exported HTML
• Keyboard-reachable: all interactive elements focusable with visible focus

Escalate shipped public surfaces to tastecheck-pass.

Invariants

Invariant floors and brief defaults live in assets/design-rails.json. Contrast, keyboard reachability, self-containment, and truthful findings stay invariant. Briefs may override measure, line-height, weight, density, texture, radius, accent count, and section rhythm only with reasons and rendered proof.

Rails contain no option lists; generate options fresh.

Difference

• v0 / LLM generators: no seed, audit, rules, reproducibility
• Coolors: random colors, no system, structure, gate
• Tasteroll: audit fixes + context-aware candidates + seeded pick + lock/reroll + complete design system + quality gate

References

• assets/design-rails.json — dimensions, floors, defaults, scope, questions, rules.
• assets/tasteroll-engine.js — pasteable xoshiro128++ PRNG with lock, reroll, shotgun, and exclude operations.

How to deliver

Deliver seed, deterministic findings, seed-dependent choices, DESIGN-SYSTEM.md, gate evidence, and design-system-interview handoff fields.

"Roll again" increments the seed; never silently reuse it with different answers.

Evidence

Contract (generated)

Canonical detail: contract.json.

• Route: Rapid design exploration or surprise direction without a full interview (+2 in contract.json); avoid: The user wants to commit to a specific direction by hand (+2 in contract.json)
• Exclude: Do not roll outside the design rails or bypass the audit findings (+3 in contract.json)
• Stop / handoff: Stop when the output will be a shipped public surface — escalate to tastecheck-pass strict gate (+1 in contract.json); receives [design-system-interview, improve-existing-website] -> sends [color-system, web-typography, spacing-system, theming, responsive-layout, art-direction, micro-motion, component-states, data-viz, empty-states, form-ux, humanize-copy]
• Output: A seeded DESIGN-SYSTEM.md with resolved findings, rolled choices, and gate evidence
• Evidence: table_with_evidence with status, reason, remediation, evidence, provenance.