We run the retrieval engine on documents you already have, and you compare the result against what your current system produces. Fifteen days. The first one is free.
A document listing, for each query you provide: which documents were returned with their exact scores, how many documents were examined, how many were provably excluded, and an Ed25519 signature over the whole record that any third party can verify without our involvement.
| Item | Detail |
|---|---|
| Corpus | Any set of documents you already search — contracts, records, decisions, policies. 1,000 to 500,000 documents. |
| Queries | Three to five questions your team actually asks. One of them should be a case where you already know the right answer. |
| Format | CSV, JSON, JSONL or plain text. One document per record. |
| Confidentiality | We sign an NDA before receiving anything. Public data works too — many teams start there. |
It does not certify that your collection was complete, that your queries covered every legal theory, or that the answer is legally correct. It certifies that the search examined every document and that nothing was skipped by heuristic. That limitation is written into the document itself — a report that promises everything is challenged in the first hearing.
Before talking to us, run it yourself on public data. The reproduction page takes two minutes and no account. The public Kaggle notebook runs the same measurement with output visible.