Field testTested AIBy DanPublished Tested Updated
ai-design-check on two of my FABLE screens: 10 warnings, 4 of them false alarms
The report
Verdict
Real pointers, plus 4 false alarms
On two of my own screens, ai-design-check named real emoji icons and real layout animations by file and line. Four of the polished page's six warnings were a matching slip that six optional semicolons cleared, with the styles unchanged.
On the tier list: One tool checked on two of my screens, with GPT-6.1 Sol running it. It places no agent on the tier list.
What the checker does
ai-design-check is a free checker from Enmera. It reads interface source code and flags habits that make AI-built screens look alike, such as emoji used as icons or animations that move the layout, and gives the file and line for each one. By the package's own description, version 0.10.0 has 87 rules; 37 of them are review guidance that a person still has to judge.
Two of my own screens
On November 2, GPT-6.1 Sol, running in Codex CLI in my workshop, ran the checker on copies of two FABLE screens: an archived older FABLE interface and the FABLE Grid design reference, my polished one. The originals were not touched. Each check took about a tenth of a second.
The first pass found nothing on the polished page. Checked as a single HTML file, it returned zero findings. Every warning after that came from the page's own styles, copied unchanged into a separate .css file beside it. If your styles live inside an HTML file, make sure the checker is actually reading them.
4 warnings and 6 warnings
The older screen got 4 warnings, all for emoji used as icons: a blueberry mark, two green check marks in status code and a moon for a category. The polished reference got 6, all for layout animation. Read as a score, the polished page looks worse. Read one by one, the warnings say something more useful.
Four warnings ran past the end of a rule
Four of the six polished-page warnings matched text that ran past the end of one CSS rule into the next. A simple background fade, for example, was read together with the next rule's border. Sol added six optional semicolons at the ends of rules, with no selector or value changed, and the count fell from 6 to 2. A headless Chromium browser read the same 303 style rules before and after.
The two warnings left point at real width, height and bottom animations. They are worth a look, but the checker measured no slowdown, and neither did this test.
A fix that did not move the count
On the older copy, Sol swapped the two green check marks for a plain ✓ and kept the blueberry and the moon on purpose. The count stayed at 4: on each of those lines, the next emoji, a stop sign and a red circle, became the reported match. A warning count is not a count of icons, so a lower number is not the goal on its own.
What else is in the video
The episode also walks through Enmera's own demos: a Tidepool dashboard going from 266 findings to 0, and a Cebbin iPhone app going from 23 to 0 over four passes, which shows the checker reading SwiftUI and Jetpack Compose code. Those are the publisher's numbers, shown as context. They were not reproduced here.
Limits of this report
- Two static screens, one run each. This gives no false-alarm rate for the checker, and the counts don't rank either design.
- Only source code was checked: no full app, navigation, accessibility or real-device test, and none of the checker's other commands, such as its narrow-layout and before/after video tools, were run.
- GPT-6.1 Sol ran the checks and made both edits in my workshop; I didn't use the tool by hand.
- The Tidepool and Cebbin counts in the video are Enmera's demos, not results from this test.
Follow the lab
New test reports and tier-list changes get announced in my Discord's #news channel. The full videos are on YouTube.
Receipts
DanManREAL · AI lab
Test receipt
@enmera/ai-design-check 0.10.0, run locally on Node 24.18.0
- Older screen, an archived FABLE interface
- 4 warnings · 0 errorsall four for emoji used as icons: a blueberry mark, two green check marks and a moon
- Polished screen, the FABLE Grid design reference
- 6 warnings · 0 errorsall six for layout animation, all in the page's styles
- Polished-page warnings that ran past a rule's end
- 4 of 6each match ran on into the next rule: background fades and a transform transition read as layout animation
- Polished copy after six optional semicolons
- 6 → 2 warningsno selector or value changed; the two left point at real width, height and bottom animations
- Same styles in a browser, before and after
- 303 rules · identicalheadless Chromium read both copies' style blocks and produced the same normalized result
- Older copy after swapping the green checks
- still 4 warningson both lines the next emoji, a stop sign and a red circle, became the reported match
- Polished page checked as one HTML file
- 0 findingsfirst pass; the six warnings came once its own style block was also saved as a .css file beside it
- Time per check
- 0.07–0.12 ssix checks across both passes and the two edited copies
- Checker that ran
- ai-design-check 0.10.0 · 37 review-only rulesthe saved copy of Enmera's code, run with Node 24.18.0; each run reported 37 rules that need a person's review
- Who ran the checks
- GPT-6.1 Sol in Codex CLI 0.166.0 · max effortNovember 2 from 9:45 AM ET; the checks ran at 9:49 and 9:50 AM
Verdict
Real pointers, plus 4 false alarms
Tier list One tool checked on two of my screens, with GPT-6.1 Sol running it. It places no agent on the tier list.
danmanreal.com/tests/ai-design-check-fable-samples
What was tested
Shorts
Related
Independent testing. Sponsorships are always disclosed, and verdicts are never for sale.
Disagree? Bring your receiptsAll reports
- One inbox for my AI coding agents: triage took 34% less time, resuming work did not improveHead-to-headNov 20
- Mythos 5.2: research and security benchmark source reviewPublic benchmarkNov 17
- Fable 5.2: agentic coding benchmark source reviewPublic benchmarkNov 17
- Claude Opus 5.5: public coding benchmark receiptsPublic benchmarkNov 17
- Astra 6.2: new model and public task-evaluation evidencePublic benchmarkNov 17
- Sol 6.2: new model and public coding benchmark evidencePublic benchmarkNov 17
- Gemini 4 Carbon: benchmark source and usage profilePublic benchmarkNov 17
- Qwen 4: Qwen4-Max launch and benchmark sourcesPublic benchmarkNov 17
- Kimi K3.1 Pro: public coding benchmark receiptsPublic benchmarkNov 17
- GLM-5.3: public coding benchmark receiptsPublic benchmarkNov 17
- Gemini 4 Argon: public coding benchmark receiptsPublic benchmarkNov 17
- Claude Sonnet 5.5: public coding benchmark receiptsPublic benchmarkNov 17
- DeepSeek V4.1 Flash: public coding benchmark receiptsPublic benchmarkNov 17
- Muse Wonder Full 1.0: public coding benchmark receiptsPublic benchmarkNov 17
- Grok 4.7: separate release and public coding resultsPublic benchmarkNov 17
- ai-design-check on two of my FABLE screens: 10 warnings, 4 of them false alarmsField testNov 2
- Vercel Agent Desktop used a sample Trip app in 6 actions without moving the mouseField testNov 1
- Cerberus Guard on sample data: one crawler name, two verdicts, and a block earned by behaviorField testOct 31
- GPT-6.1 Terra vs Opus 5.5: three real projects, and the designs I picked blindHead-to-headOct 29
- Steps → Reduced: opening a terminal agent went from 98 seconds to 11AuditOct 28
- One ChatGPT chat directed Opus and Codex at once through DMR_REMOTEField testOct 28
- Claude Duet ran my channel for a week and caught 269 fake AI crawlersField testOct 27
- Kimi K3.1 Pro built my animation family, after a quota stop and 21 minutes of failed requestsField testOct 24
- ChatGPT + Codex built 33 new DMR_REMOTE toolsField testOct 23
- Gemini 4 Argon reviewed a finished episode: 14 clips, one flag, one inconsistencyField testOct 21
- Qwen 3.8 Flash Next on my RTX 5080: 81 tokens a second with 202K tokens readAuditOct 21
- Does an extra AI review step pay off? A small DeepSeek trial said noHead-to-headOct 19
- My real DeepSeek bill: 12.31 billion tokens for $122.28Cost receiptOct 17
- Gemini 4 Argon kept hanging in Gemini CLI. The cause was a 60-second timeoutIncidentOct 14
- I asked 10 AI coding agents for their biggest weaknessSelf-reportOct 14
- Jarvis 27B on my own PC: 30 tokens a second, and it still recalls at 251KAuditOct 10
- My agents' first game benchmark was wrong. Here's how it was caughtIncidentOct 10
- Pixal3D + Blender turned my 2D art into animated 3D charactersField testSep 22
- DeepSeek Flash answered 200 OK and sent nothing for two hoursIncidentSep 14
- Grok, Codex and DeepSeek played WorldOS for Rank 1. Grok won with 98Head-to-headAug 20
- DeepSeek V4 Flash audited 18 of my real coding sessions. Two said 'done' too earlyAuditAug 4