24 real documents, and exactly what happened to each.

The three statuses

PASS

Fixed, validated, delivered. Charged only at this point, never before.

REFUSED

We attempted it and our own gates rejected the result. Nothing delivered, nothing charged — this is the guarantee firing.

OUT-OF-SCOPE

Not work this service does — a scan with no text, a file we can't open, a document already tagged by someone else. Routed before any processing, never charged.

Reason codes

PASS — 15 documents

Each passed all seven structural checks, passed full PDF/UA-1 validation, and kept at least 91% of its text reachable by a screen reader. That last number is ours, not the standard's: a document can pass ISO validation with most of its text hidden from assistive technology, so we measure exposure separately and refuse to deliver a file that reads as mostly empty.

REFUSED — 1 document

Attempted, rejected by our own gates, not delivered, not charged.

OUT-OF-SCOPE — 8 documents

Recognised and routed rather than attempted. A scan has no text to tag; an already-tagged file needs repair, not rebuilding, and we never overwrite someone's structure blindly; an unopenable file is unopenable. The free checker tells you which of these you have before you upload anything.

Method

Each document runs through the identical pipeline customers get, then is validated with veraPDF, the open-source ISO 14289-1 reference validator. Documents that fail any gate are refused and free — refusals below are the guarantee working, not failures of it.

• Documents come from public sources — arXiv, the IRS, NIST, the UN, the Federal Reserve, Berkshire Hathaway — plus constructed edge cases (a scan with no text layer, a truncated file, a password-protected file, an oversized document) that exist to test refusals rather than successes.
• Every file runs through the same code path a paying customer's document takes, with every gate enabled. No hand-tuning, no per-document exceptions.
• Validation is performed by veraPDF, which we don't control, and the version is recorded above. The same validator is what compliance auditors use.
• Statuses and codes are assigned by the same script that reads the run's raw results, so this table cannot say something the test run didn't measure.
• This is a sample, not a complete history — we run new documents regularly and publish a run once, in full. If you want a class of document tested that isn't represented here, send it and we'll add the result whichever way it goes.

Two honest limits: conformance to a named validator is not a legal opinion, and no machine can judge whether an image description is well written — validators check that alt text exists, not that it's good. We don't claim otherwise anywhere.