every line pdf-to-md audit prints, in order: one per page with
missing words, then summaryLine
words genuinely absent from the markdown after repair-checking; 0 is
the only passing value (pdf-to-md audit exits nonzero otherwise)
lines in the dump that were not parseable JSON; auditWords rejects evidence with any such line
distinct page numbers present in the dump
JSON lines successfully parsed from the word dump
per-page detail for every page with at least one missing word (capped sample of the missing tokens per page)
words present in the dump but not found verbatim on their page, BEFORE
hyphen-join/split repair is applied; always >= missing
the one-line parsed=… pages=… words=… strict_deficits=… MISSING=…
summary, also the last entry of lines
total recognized words across all pages in the dump
auditWords()'s result — the severity-1 no-silent-loss check.