New ML-Powered Contrast Checker Reduces False Positives by Learning Typographic Context

AI · 3 min read

The new tool departs from strict ratio thresholds and uses a model trained on human perceptual judgements across typography, language, and viewing distance. It factors in font x-height, stroke contrast, diacritics density, and typical UI scaling to produce a context-aware accessibility score. Early users report it is better at differentiating problematic combinations—like high-contrast but low-legibility pairings—from acceptable variance in decorative contexts.

Offered as both an API for CI integration and a Figma plugin for designers, the product surfaces inline recommendations: swap to a more legible font weight, slightly increase tracking, or substitute a token that preserves brand color while improving legibility. The startup emphasizes that the model complements, not replaces, WCAG checks: it flags areas where designers might prefer human review instead of strict pass/fail logic.

Accessibility consultants welcome the nuance but caution against relying solely on ML judgment for compliance. The company counters by providing an audit trail—each recommendation links back to measurable typographic parameters and offers toggles to revert to strict WCAG evaluation when legal or procurement requirements demand unequivocal metrics.