Document conformance, claim by claim.
Cited, or silent.
Docyma breaks a document into individual claims and checks each one against the text of a regulation. Not a summary: the text itself, pinned to a version. A finding must quote the passage it stands on, word for word, linked into EUR-Lex at paragraph level. No quote, no finding. First corpus: the EU AI Act.
Cited or silent
Every finding carries the exact words of the regulation it rests on, and the quote is checked mechanically against the source before it reaches the report. Where the corpus says nothing, Docyma says nothing.
Same document, same report
A corpus is a pinned snapshot of the source text, hash and all. Run the same document against the same corpus version a year from now and you get the same report. It names the version it was built from.
Every finding is cross-examined
Before an incongruence reaches the report, a second model argues against it. What survives keeps the trace of that review. A false accusation is the expensive failure: the pipeline is built around that.
Docyma is in active development and already runs end-to-end on our own test corpus. The EU AI Act comes first; GDPR, ESRS and public procurement next. The engine doesn't care which regulation it reads.