Check every source locator in a record against the stored page: each extracted value's UTF-8 byte range is cut from the sealed page bytes (re-hashed first) and compared with the recorded value. With finding=CODE, the same for a finding: the robots.txt lines, the HTML tag served where a file should be, the empty <urlset>, an absence re-checked over the whole stored file, a manifest field

one fact, e.g. title or telephone

a finding code; without a licence only the top finding

name, phone, locality, experience or all: both sides of a claim (what the structured data asserts, where the visible page says it) cut from the stored page, plus the rule that judged it, the inputs it read, the row it decided, and that rule rerun over the stored page (replay.reproduces); points per row with a licence or the owner's session