WHOSEARCH / TRUST CENTER

People search benchmark protocol

Accuracy numbers will be published only after a fixed test set is run. This protocol defines the measures in advance.

Query planning

Searches combine the user input with optional public context. The system does not infer hidden attributes or use leaked credentials.

Candidate separation

Namesakes and reused handles remain separate unless multiple public anchors support the same identity.

Source handling

Each result keeps its original URL, visible excerpt, source domain, and check time. A missing page is not replaced with an invented fact.

Uncertainty and corrections

Conflicts, missing clues, and weak matches remain visible. Users can report a source or request correction and removal through support.

Measures fixed before testing

Top-1 and Top-3 correct match rate, false match rate, supported-source rate, coverage, abstention rate, time to first verifiable source, and search cost.

Current status

The protocol is published. No accuracy percentage is claimed until the representative test set and independent review are complete.