Checking that a quote is in the page it cites
A quotation nobody verified is a quotation somebody invented.
What it is
Models fabricate quotes. Not often, and not obviously — they produce something the review could plausibly have said, attached to a real URL. It is the single most damaging thing a report of this kind can contain, because it is the part a customer is most likely to forward.
So every quoted string is searched for in the text of the page it cites. A quote that is not found is not deleted — the finding survives, the quotation is demoted to a description, and the report says which happened. Three outcomes, not two: found, not found, and the page carried no text we could check.
How far it goes
Quotes shorter than 12 characters are not checkable and are not treated as quotes.
Performed by
research/schema.py. This page names the functions that do it, and
the build fails if any of them is renamed — marketing copy has no
compiler, so we gave this page one.