Find the duplicate people in a messy list

Records that disagree with each other — married names, nicknames, initials, half-filled rows — go in. What comes back is the merges the engine is confident about, the pairs it is confident are different people, and the short ranked list a human should actually look at. Every decision carries the evidence that produced it — an audit record you can check, not a score you have to trust. Read the names and judge it yourself; we have not marked our own answers.

Pasted records are resolved in memory and never stored — the same no-retention promise the API makes about everything you send it. The whole policy is four paragraphs.

Worked on your list? Get a key — 1,000 free credits, no card, no sales call — and the paste box’s cap rises from 100 records to 5,000.

These people are invented. The failure patterns are not — each one is a category measured against 7.5M North Carolina voters, 170k ORCID researchers and 755k OpenSanctions analyst judgements. Those records are real people and never leave the machines that hold them, so the patterns travel here and the people do not. That makes this an honest demonstration of behaviour, not an accuracy claim: this list was generated by the same project that wrote the matching rules. The numbers that mean something — measured against those corpora, including what the engine still gets wrong — are written up separately; ask for them at casey@sameornot.com.

This is the POST /v1/person/batch endpoint, unmodified — up to 5,000 records, synchronous, nothing stored. Read the API docs.