Nothing you upload is stored, logged, or sent anywhere. The PDF is parsed in memory and forgotten once the response is returned.
Best-effort text-layout parsing of a real document, not a guaranteed-correct one: this was built and tested against DLA's own IC PDFs (which share one consistent generator format). A convention from a different publisher or generator may parse only partially, or not at all -- rows that don't match an expected shape are skipped rather than guessed at.