Taste Audit Scorecard (Printable)

What this is

A one-page review scorecard for any AI-generated design. You can use it in three ways:

1. Solo review — print it or copy the markdown, fill it out by hand after each generation
2. Team review — share it in Notion or Figma as a review artifact
3. AI review — paste it into a Claude Code skill (see the Taste Skill Starter)

The scorecard is meant to be repeated. Every generation gets scored. Over time, you see which dimensions stay weak and you tighten your CLAUDE.md rules to fix them.

---

The scorecard

# Taste Audit — [PROJECT / PAGE NAME]

**Date:** [DATE]
**Reviewer:** [YOU]
**Generation tool:** [Claude Code / Cursor / Lovable / v0 / other]
**File or URL:** [PATH]

---

## Scores (0–10)

| Dimension | Score | Notes |
|-----------|-------|-------|
| 1. Color discipline | ___ / 10 | Every color from the approved palette? |
| 2. Typography | ___ / 10 | Max 3 sizes, max 2 weights, correct font per role? |
| 3. Spacing rhythm | ___ / 10 | Values from the spacing scale only? |
| 4. Motion restraint | ___ / 10 | Stayed at the approved motion level? |
| 5. Copy tone | ___ / 10 | Correct person, no buzzwords, statements not teases? |
| 6. Anti-pattern cleanliness | ___ / 10 | Zero hits on the 10 anti-patterns? |
| 7. Authorship | ___ / 10 | At least one element feels specifically authored? |

**Total: ___ / 70**

Pass threshold: 56 / 70 (every dimension at 8+ AND zero anti-rule hits)

---

## The 10 anti-patterns (check all that apply)

- [ ] 1. Luxury by default (dark + gold + serif + "premium" language)
- [ ] 2. Gradient overload (>1 gradient, glow effects, colored shadows)
- [ ] 3. Fake depth (default shadows on cards, glassmorphism, stacked shadows)
- [ ] 4. Random 3D hero (abstract spheres, orbs, blobs, mesh gradients)
- [ ] 5. Over-animated (scroll-triggered, parallax, scale-on-hover, bounce)
- [ ] 6. Spacing without rhythm (values outside the scale)
- [ ] 7. Typography without hierarchy (>3 sizes, >2 weights, all-similar headlines)
- [ ] 8. Reference soup (Apple + Stripe + Linear + luxury on one page)
- [ ] 9. No content logic (buzzwords, adjective-only feature cards)
- [ ] 10. Generic polish with no worldview (fine but forgettable)

Any box checked is an automatic fail, regardless of scores.

---

## Violations and fixes

| Violation | Where | Fix applied |
|-----------|-------|-------------|
| | | |
| | | |
| | | |

---

## Taste-stack updates triggered by this review

After the audit, update your taste stack if needed:

- [ ] Added rule to project `CLAUDE.md`: __________________________
- [ ] Added rule to personal `~/.claude/skills/taste/SKILL.md`: __________________________
- [ ] Added anti-rule to personal taste skill: __________________________
- [ ] Updated `/anti-pattern-check` skill: __________________________

---

## Decision

- [ ] **Ship.** Scores and anti-patterns pass. Ready to release.
- [ ] **Fix and reship.** Violations fixed in this review, re-audit scheduled.
- [ ] **Reject.** Output needs regeneration with tightened rules.

---

## Notes

[Free text space for anything that did not fit the scorecard]

---

How to score each dimension

1. Color discipline (0–10)

• 10: Every color on the page is from the approved palette. Zero exceptions.
• 8: One unauthorized color on a low-priority element (caption, border).
• 5: Multiple unauthorized colors, OR the accent is used in a forbidden way.
• 0: The palette has been ignored. New colors appeared without approval.

2. Typography (0–10)

• 10: Correct fonts, max 3 sizes, max 2 weights, perfect hierarchy.
• 8: Minor hierarchy issue (one heading slightly wrong size).
• 5: Wrong font used, or more than 3 sizes.
• 0: Generic system font applied, or chaotic size mixing.

3. Spacing rhythm (0–10)

• 10: Every padding and gap value is on the spacing scale.
• 8: One or two values slightly off-scale.
• 5: Multiple off-scale values, section spacing inconsistent.
• 0: Spacing appears randomized, no rhythm.

4. Motion restraint (0–10)

• 10: Output stays at the approved motion level. No surprises.
• 8: One motion slightly too much (e.g. 200ms instead of 150ms).
• 5: Multiple violations (scroll-triggered reveals, scale-on-hover).
• 0: Motion spam. Parallax, bounce, stagger, everywhere.

5. Copy tone (0–10)

• 10: Voice matches rules perfectly. No buzzwords. Headlines are statements.
• 8: One buzzword or one wrong headline style.
• 5: Multiple buzzwords, generic SaaS copy creeping in.
• 0: Copy is pure SaaS boilerplate. “Revolutionize your workflow.”

6. Anti-pattern cleanliness (0–10)

• 10: Zero anti-patterns from the list of 10.
• 8: One borderline anti-pattern, easily fixed.
• 5: Two or three anti-patterns present.
• 0: More than three anti-patterns. Output reads as generic AI.

7. Authorship (0–10)

• 10: At least one element is clearly, specifically authored. The page would be recognizable as yours.
• 8: An authored element exists but is subtle.
• 5: The page is competent but could belong to anyone.
• 0: Pure SaaS template. No worldview at all.

---

When to use the scorecard

• After every AI generation if you are still tuning your CLAUDE.md
• Before shipping any client-facing page
• During team review to create a shared review language
• As a portfolio artifact showing how you evaluate AI output

---

How to improve scores over time

1. After each audit, find the 1–2 lowest-scoring dimensions.
2. Look at the specific violations in that dimension.
3. Add a rule to your CLAUDE.md or taste skill that would have prevented those violations.
4. Regenerate and re-audit. Score should rise.
5. Repeat.

Within a few audit cycles, your generations should consistently score 56+ without manual fixing. That is the point where the taste stack has matured.

Pairs with

• The AI Anti-Patterns Gallery (detailed description of the 10 anti-patterns)
• The Build Your Own Taste Skill workshop (turns this scorecard into an AI-driven audit)
• The Anti-Pattern Check Skill template (runs the anti-patterns check automatically)
• The Motion Audit Skill template (runs the motion check automatically)