Connections should be inspectable, not magical.
Anomaly Graph is designed to help people find relationships while keeping the underlying record, its provenance and its limitations in view.
1. Keep evidence classes separate
Government case files, archival passages, historical or religious texts and completed research dossiers can all be searched, but they are not treated as interchangeable. Each result carries its source family and the best available route back to the underlying material.
2. Search the words people actually use
Users can enter a word, an exact phrase or comma-separated terms. Comma stacking asks for records containing every supplied term. Related-language expansion can surface nearby wording, but the interface distinguishes an exact passage from a related lead.
3. Preserve provenance
Source links, page references, extraction confidence and review status are retained wherever available. OCR may be imperfect. Historical translations and dates may be disputed. Those limitations travel with the result rather than disappearing behind a summary.
4. Label AI-sourced material honestly
Dossiers marked “Sourced using AI” are discovery material assembled from model knowledge. They are not documentary evidence and must not be presented as independent corroboration. Important claims should be checked against retrieved primary or reliable secondary sources.
5. Treat patterns as questions
A resemblance can justify further research. It does not establish identity, causation, independence or an extraordinary explanation. Anomaly Graph helps formulate the next question; it does not manufacture a conclusion.