Accuracy you can check.
Every release of OTTO sits an exam: 57 audited questions, written to be hard, over the full library. Five gates, five pass bars. No release ships without passing all five. Measured 25 August 2026.
Cites the right source
When an answer points at a document, it is the correct document, section and paragraph.
Pass bar 95%
Claims grounded in the documents
Every statement is supported by the quoted passages, checked independently against the passages the answer cites.
Pass bar 95%
Finds the right legal basis
The passages an answer needs are actually retrieved, including when they live in statute rather than policy.
Pass bar 90%
Declines unanswerable questions
When the documents do not answer, OTTO says so rather than guessing.
Pass bar 90%
Catches trick questions
Questions built around common misconceptions and out-of-date law get the substantively correct answer.
Pass bar 90%
Why you can trust a score we publish ourselves.
Because you do not have to trust it. Every answer carries verbatim quoted passages with document and paragraph references, and the source documents are public: pick any answer and mark it yourself. The exam includes questions the documents genuinely cannot answer, and declining to guess is scored as strictly as answering correctly. Results are published in full whether a run passes or fails.