Human-verified checks

What the rule requires

Some WCAG criteria are genuinely judgment: does an instruction rely on shape or position? Is text living inside an image? Is focus visible? No honest tool calls these automatic.

Proof’s answer is structural: machine-detected candidates, evidence on the review card, and a named human attestation recorded on the certificate. The claim is automated + attested — never “fully automatic.”

• Sensory characteristics1.3.3 · A
• Resize text1.4.4 · AA
• Images of text1.4.5 · AA
• Non-text contrast1.4.11 · AA
• Keyboard access & traps2.1.1 / 2.1.2 · A
• Timing & flicker2.2.1 / 2.3.1 · A
• Focus visible2.4.7 · AA
• Language of parts3.1.2 · AA

How Proof clears it

1. 01

   Proof detects the candidates: bilingual scans for instructions that rely on shape, position, or sound (deliberately over-flagging for the reviewer), OCR-diff evidence for images of text, Unicode-script scans for language of parts, form-control border measurement for non-text contrast, and script and timing heuristics for the rest.

2. 02

   A named human attests, with the evidence beside the decision. The attestation is attributed, and it flows onto the compliance certificate.

3. 03

   The gate holds: an attestation can clear a review item, but it can never override a confirmed failure.

What to expect

Criteria that don’t apply are honestly N/A — a document with no bordered form controls, no scripts, and no media isn’t padded with vacuous green checks. And unverified human checks read “needs review”, never “failed”: that word is reserved for confirmed violations.