
AI color palette: Cyan gradient background
div.hero
- snippet
- Cyan gradient background
- name
- AI color palette
- impeccable category
- slop
D:\dev\projects\restyle\slopbench\corpus\s04\index.html
Tells 27 → 0, second-order 0 → 0, distinctiveness 0.62 → 0.40 (worse), judge preferred after (blinded, confidence 2/3).
Worse after the redesign: LCP 1.57 s → 1.58 s; Distinctiveness 0.62 → 0.40.
Of 15 directional metrics measured on both sides, 7 improved, 6 stayed the same and 2 got worse.
Every metric before and after, judged per metric. Missing numbers say so.
| Metric | Before | After | Change | Result |
|---|---|---|---|---|
| Tells | ||||
| Tells, total | 27 | 0 | −27 | Better |
| High severity | 1 | 0 | −1 | Better |
| Second-order clichés | 0 | 0 | 0 | Same |
| Accessibility | ||||
| axe serious + critical | 1 | 0 | −1 | Better |
| Contrast failures | 11 | 0 | −11 | Better |
| Lighthouse (lab, mobile) | ||||
| Performance | 99 | 99 | 0 | Same |
| Accessibility | 94 | 100 | +6 | Better |
| Best practices | 96 | 96 | 0 | Same |
| SEO | 100 | 100 | 0 | Same |
| LCP | 1.57 s | 1.58 s | +7 ms | Worse |
| CLS | 0.049 | 0.000 | −0.049 | Better |
| TBT | 0 ms | 0 ms | 0 | Same |
| Typography | ||||
| Distinct font sizes | 16 | 7 | −9 | Changed |
| Size entropy | 3.15 bits | 1.97 bits | −1.18 bits | Changed |
| Palette | ||||
| Hue entropy | 1.30 bits | 0.52 bits | −0.78 bits | Changed |
| Chromatic colours | 3 | 1 | −2 | Changed |
| Model-based | ||||
| Distinctiveness | 0.62 | 0.40 | −0.22 | Worse |
| Prototypicality, structural | 0.88 | 0.88 | 0 | Same |
| Prototypicality, niche match | 0.14 | 0.32 | +0.18 | Better |
| Convention | Before | After |
|---|---|---|
| header-nav-at-top | kept | kept |
| brand-in-header | kept | kept |
| h1-above-fold | kept | kept |
| cta-above-fold | kept | kept |
| contact-or-pricing | kept | kept |
| footer-with-links | missing | missing |
| mobile-no-horizontal-scroll | kept | kept |
| mobile-readable-single-column | kept | kept |
Above the fold at each width. The redesign has the dark frame.
Each finding with its rule, element and measured evidence, and what happened to it.
33 findings before, 0 after: 33 fixed, 0 still present, 0 new. Findings are matched by id, then by rule and selector.
Every finding in these categories is gone from the after audit. Open one to see the original evidence.

div.hero

body

body

p.s

#services > div.wrap > p.sub

div.card:nth-of-type(2) > p

li:nth-of-type(2) > p

p.note

#how > div.wrap > p.sub

div.wrap.price > div:nth-of-type(1) > p.sub

div.card:nth-of-type(1) > h3

div.card:nth-of-type(2) > h3

div.card:nth-of-type(3) > h3

div.card:nth-of-type(4) > h3

div.card:nth-of-type(5) > h3

div.card:nth-of-type(6) > h3

#services > div > div:nth-of-type(2)

#services > div > div:nth-of-type(2)

body

body > header > div > nav > a:nth-of-type(5)

#services > div.wrap > div.eyebrow

div.wrap.price > div:nth-of-type(1) > div.eyebrow

section:nth-of-type(5) > div.wrap > div.eyebrow

#faq > div.wrap > div:nth-of-type(1) > div.eyebrow

#how > div.wrap > div.eyebrow
![Before, 1440px, crop of nav > .btn-alert.btn[href="tel:+15550142277"]](assets/before/crops/7dbbdb3199.webp)
nav > .btn-alert.btn[href="tel:+15550142277"]

#services > .wrap > .eyebrow

.price > div:nth-child(1) > .eyebrow

section:nth-child(6) > .wrap > .eyebrow

#faq > .wrap > div:nth-child(1) > .eyebrow

.final > .wrap > p

nav > a.btn.btn-alert

#services > div.wrap > div.eyebrow
A model compared the two versions without knowing which was which.
The judge saw two unlabeled screenshots, A and B, at 1440 and 375 px in random order, and never knew which was the redesign; the labels were mapped back afterwards.
B makes the emergency number itself the largest element on the first screen at both widths with orange kept for call actions, while A leads with a two-colour slogan over a template-style gradient hero with a pill badge and a curved divider, and on phone its header call button wraps onto two lines and its sticky bar covers the callback button.
| Criterion (1-5) | Before | After | Change |
|---|---|---|---|
| Looks generic-AI | 2 | 4 | +2 |
| Hierarchy | 3 | 5 | +2 |
| Typography | 3 | 4 | +1 |
| Colour | 3 | 4 | +1 |
| Fit to niche | 4 | 5 | +1 |
| Trust | 3.5 | 4 | +0.5 |
How the numbers were produced, and what they cannot tell you.
| Tool | Version | Status | Note |
|---|---|---|---|
| playwright | 1.63.0 | ran | chromium 153.0.8010.12 |
| axe-core | 4.14.1 | ran | 1 violations, 17 passes; 0 violations, 25 passes |
| impeccable | 4.10.0 | ran | cdp: 22 findings at 1440px; cdp: 20 findings at 375px; cdp: 0 findings at 1440px; cdp: 0 findings at 375px |
| lighthouse | 13.6.0 | ran | mobile, simulated throttling |
Distinctiveness embeds the page's screenshots with an image model and measures how far they sit from a reference set of landing pages generated by AI coding tools with default prompts. It is reported as 0 to 1, where 1 means further from every reference page. This run used the refset reference set and the Xenova/clip-vit-base-patch32@q8 embedding model. Before the redesign, the closest reference was s16 at similarity 0.86.
Prototypicality has two parts. The structural score is the share of this niche's expected conventions the page keeps, checked in the DOM. The niche match is a zero-shot image classification: the probability that the screenshot reads as its stated niche rather than a neighbouring one. Visitors trust pages that look like their category, so a redesign should be distinctive without losing these.