Every row is one extracted incident. Titles, shapes, and behaviour are our normalized analysis; each links to a record page showing the verbatim claim and the source document. The legibility bars measure how cleanly the fields could be extracted from the source — not the credibility of the sighting; low bars are noisy mid-century scans and heavily-redacted compilations.