Methodology
Not a black box.
Every score has a source.
Most AI tools give you a number. Corexi gives you the number, the research it comes from, the evidence behind it, and the code to fix it. Here is exactly how we measure product experience.
01 — The score
One number. Five dimensions. Weighted.
Your UX Score is a weighted composite of five UX dimensions. Each dimension is derived from 9 core evaluation categories plus 2 neurodiversity layers. The formula is public, not hidden.
Multi-viewport: desktop carries 60%, mobile 40%. Each scan accumulates product context, so scores get sharper over time.
02 — What we measure
9 core categories
Every page scan is evaluated across these categories. Each has measurable criteria rooted in 200+ peer-reviewed sources. Every accessibility finding is cross-checked against axe-core.
Visual Hierarchy
Is the most important element the first to catch attention? Are headings differentiated? Is the visual flow logical?
CTA Visibility
Can you find the primary action in under 2 seconds? Is contrast above WCAG AA 4.5:1? Is sizing appropriate?
Form Design
Are labels present and readable? Are error states designed? Are inputs sized for touch and keyboard?
Spacing & Layout
Is white space consistent? Is a grid system present? Are there cramped or scattered areas?
Typography
Are fonts readable at 16px+ body? Is line length 45-75 characters? Does weight create hierarchy?
Color & Contrast
Does text pass WCAG AA? Is the palette consistent? Is color ever the sole information carrier?
Mobile Responsiveness
Are touch targets 44x44px+? Is there horizontal scroll? Is text readable on a phone?
Cognitive Load
Can a first-time user understand what to do? Is navigation clear? Is there decision paralysis?
Accessibility
Alt text, form labels, keyboard nav, ARIA, focus indicators. Every claim axe-core verified.
03 — Beyond standard UX
+2 neurodiversity categories
WCAG covers compliance. It does not cover whether a neurodivergent user can actually use your product. Corexi scores Color Vision Deficiency and Dyslexia Readability as first-class scan categories. No other automated tool does this.
Color Vision Deficiency
Red/green combinations, color as sole info carrier, problematic color pairs. Scored as a dedicated scan category.
Dyslexia Readability
Line length, letter spacing, justified text, ALL CAPS blocks, contrast below 7:1. Scored as a dedicated scan category.
04 — Real user data
Your analytics + our snippet. One confidence score.
Visual analysis tells you what looks wrong. Behavioral data tells you what actually hurts. Connect your existing analytics — or drop our one-line Corexi Snippet for instant signals without any third-party dependency. The more sources you connect, the sharper the findings.
| Provider | Window | Signal |
|---|---|---|
| GA4 | 7 days | Sessions, bounce rate, engagement |
| Clarity | 72 hours | Rage clicks, dead clicks, quick-backs |
| Mixpanel | 7 days | Event volume, active days |
| Amplitude | 7 days | Active users, event cadence |
| PostHog | 7 days | Events, active users (deduped) |
| Firebase | Auth | Mobile app analytics |
| Hotjar | Auth | Survey and feedback data |
| Corexi Snippet | 1 hour | Scroll depth, rage clicks, dead clicks, time on page |
~40%
Visual AI only
~65%
+1 provider
85-95%
+2-3 providers
05 — Verification
Verified, not guessed.
Every accessibility finding is cross-checked against axe-core at runtime. Verified findings carry a badge. LLM-only findings stay flagged. You always know which is which.
Verified finding
AI detected an issue AND axe-core confirmed it against WCAG 2.2 rules. 27 rule IDs checked. This is the standard for compliance claims.
AI-only finding
AI detected a potential issue but no deterministic rule confirmed it. Flagged for human review. Transparent about confidence level.
06 — The layers
Three layers. One score.
Every scan flows through three analysis layers. Each layer adds a different kind of evidence. Together, they produce findings that no single layer could generate alone.
Visual layer
Multi-viewport screenshots analyzed by AI vision across all 11 categories. Deterministic checks (contrast ratios, touch targets, spacing) run alongside AI interpretation. This is what you see.
Behavioral layer
Real user signals from your connected analytics and the Corexi Snippet — bounce rates, rage clicks, scroll depth, session patterns. This is what your users actually do.
Reasoning engine
Cross-validates visual findings with behavioral data. A contrast issue that correlates with high bounce rate scores higher than one with no behavioral signal. The engine gets sharper with every scan.
Visual-only scans work from day one. Adding behavioral sources increases confidence from ~40% to 85%+. The engine never requires all layers — it adapts to what you connect.
07 — Data principles
Privacy-first. Always.
Minimal, opt-in tracking
The Corexi Snippet is optional and cookieless. It collects anonymous behavioral signals only — no PII, no session replay, 24h data retention. Your users see nothing.
Read-only analytics
We connect to your existing providers via read-only API. We pull aggregated metrics only.
No PII, ever
Bounce rates, session depth, engagement patterns. Never individual identities, emails, or browsing histories.
EU-based, GDPR compliant
All processing on EU infrastructure. Credentials encrypted at rest (AES-256). Disconnect and delete any time.
See it in action.
Book a demo and see how the engine works on your own product.