VIBEBRATOR
Take the Vibe Check
The Bench

We ran our own audit on this site

2 September 2026 · full site · no mercy mode · second pass same day

61%
Challenger

Good bones. Several Giants would reject this. Below is the whole thing — scorecard, three ship-blockers, and what we did about them.

Every resource site tells you to hold your code to a standard. This is the only one we know of that publishes the standard, runs it on itself, and shows the result including the parts that fail. A site scoring itself 95% would tell you nothing. This one scored 61%.

What this is

The Titan Audit scores a codebase against the qualities of the most valuable companies on earth — each mapped to something testable, not a metaphor. You can run it on your own project from the Prompt Deck on the home page, in four modes.

What we measured

MetricValue
Total payload234 KB across 6 pages, static, no backend
Largest fileindex.html — 2,956 lines
Test files0
AnalyticsNone loaded
Content visible without JavaScript11% — 975 of 8,860 characters
Secrets in sourceNone

Scorecard

GiantAvgCritical finding
Disney8.3Nothing. Best dimension on the site.
Ferrari8.0Distinctive and hand-made — but JS-gated
Home Depot7.3No contributor docs, no error copy
AMD7.3Graceful degradation genuinely well done
Bank of America7.034 touch targets under 44px; no skip link
Microsoft6.3No build step, no automation
Netflix6.0Nothing personalised; no data to do it with
Palantir6.0Cannot answer “what happened”
Apple5.72,956-line file; tokens duplicated six times
Amazon5.7No error or empty states
Berkshire5.7One design system copy-pasted into six files
JPMorgan5.3No revenue mechanism is actually live
Johnson & Johnson4.7Privacy policy described features that did not exist
NASA4.3No metrics, no flags, no way to know what works
Eli Lilly3.7Zero tests. Zero monitoring.

The three ship-blockers

P089% of the site was invisible to AI crawlers

Measured: 975 characters rendered without JavaScript, 8,860 with. The toolbox, the prompt deck and the word of the day were all built client-side. Google renders JavaScript. GPTBot, ClaudeBot and PerplexityBot largely do not. A site whose whole argument is that it deserves to be cited had built itself so the citing systems could not read it.

Fixed. There is now a build step that pre-renders every section into the HTML. Content visible without JavaScript went from 11% to 100%. There is also an llms.txt carrying the entire site as plain text.

P0The privacy policy described things that did not exist

It claimed outbound clicks were counted through a /go/ redirect — there was no such route. It claimed aggregate analytics — none were installed. It promised a consent banner — there was no consent code on any page. A privacy policy describing processing you do not perform is a false statement in a legal document.

Fixed. The policy now says plainly that the site sets no cookies, runs no advertising, and has no affiliate links, and describes what will change before any of that starts.

P0The only contact address did not exist

Five mailto: links across three legal pages pointed at a mailbox that had never been created. A privacy contact route that bounces is worse than none.

Fixed. Mailbox created and MX/SPF records live; a test message from an outside account was received on 2 September 2026.

Closed since the audit

Same day, second pass. Each line below is checked by a script that fails the build if it stops being true (test.sh in the repo: build gate, unit tests, static checks, a headless browser with axe, and an end-to-end run of the Vibe Check).

FindingWasNowChecked by
Tests08 unit + 22 end-to-end assertions + 9 pages of static and browser checkstest.sh
Text visible without JS11%98–113% on every page (measured innerText, JS off vs on)site_check.py
Design tokenscopied into 6 filesone file, chrome.css; a page that defines its own :root fails the buildstatic check
Header3 different headersone header, one footer, generated for all 9 pages; nav text compared across pages in a browserbrowser check
Canonical URLapex, while apex redirects to wwwhttps://www.vibebrator.com/… everywhere: canonical, og:url, sitemap, robots, llms.txt, JSON-LDstatic check
Header CTApointed at a newsletter that does not existpoints at the Vibe Check, which does246 internal links resolved
Affiliate disclosure“Some links pay us” while the policy said none doremoved; the policy and the page agreestatic check
Hero counters“142 prompts”, inventedderived from the data files at build time (20 prompts, 95 tools, 60 terms)build
Toolbox duplicates“Hugging Face” ×4, “McStumble” ×2named for what they are; duplicate names or slugs fail the buildstatic check
Legal pagesdifferent header, no og:image/og:url/twitter cardsame chrome and full metadata as every other pagestatic check
/stack without JSfour dashes where the text should berebuilt as a vector building; every floor and card is in the markup, 100% visible, fully usable without the animation librarybrowser check + 36 e2e assertions
axe (WCAG 2.1 AA)not run0 violations on 9 pages; card contrast and a nested-interactive SVG were fixed to get thereaxe-core 4.10
Contact mailboxdid not existhello@vibebrator.com receives mail (MX + SPF verified via public DNS, test message delivered)manual, 2 Sept
Console errorsnot measured0 on 9 pages with the CDN blockedbrowser check

Still open

P1Zero monitoring. If the deck breaks for every Safari user tomorrow, we find out when someone tells us. Adding it means changing the privacy policy first.

P1No revenue mechanism is live. Two ad slots are reserved in the homepage markup and switched off. They stay off until a consent platform, a truthful policy, and the mailbox exist, in that order.

P2Interactive elements under 44px. Every control now clears the 24px WCAG 2.5.8 floor; 6 to 40 per page are still under the 44px target, mostly filter pills.

P2The homepage file is large — 330 KB before compression, because 95 wireframe previews are inlined as SVG so the page works without JavaScript. Compressed on the wire it is a fraction of that; it is still the next thing to measure.

What scored well

Graceful degradation. Every animation is guarded; the page is fully functional with the animation CDN blocked, and a three-second safety net force-shows anything a failed observer left hidden.

Contrast. Five of six colour tokens pass WCAG AA, four above 8:1. The sixth measured 4.14:1 and was lifted.

Accessibility fundamentals. Landmarks, one h1, language set, zero unlabelled buttons, zero unlabelled SVGs, zero images without alt text.

No attack surface. No backend, no auth, no user input reaching anything.

Why publish a 61%

Because the alternative is asking you to verify ruthlessly while hiding our own numbers. The gap between a site that says “hold your code to a standard” and one that shows its own score is the entire difference between a content farm and something worth reading.

Next target: 78% — Contender. We will publish that one too, including whatever it finds.

Run the audit on your own project →