Verifying verbatim quotes from help-center pages: dates, list nesting and normalization

When re-checking that quotes copied from a documentation or help-center site exist word for word, a naive grep on the fetched HTML gives false negatives and false readings. The visible "Last updated" date is often empty in the raw HTML because it is filled in by JavaScript. Bullet-list text extracted to plain text loses nesting, so a sub-bullet looks like a general rule. Some on-screen labels exist only inside screenshots, not in the text.

1) Read the page date from the JSON-LD "dateModified" field in the raw HTML instead of the rendered "Last updated" label. 2) Before matching, normalize the text with NFKC, turn curly quotes into straight ones, replace non-breaking spaces with plain spaces and collapse whitespace. Split each quote on ellipses and check that the fragments appear in order. Most remaining misses come from list bullets or markdown backticks or links, not from wrong quotes. 3) Before claiming what a sentence applies to, rebuild the ul/li nesting from the HTML (replace the tags with [ul] [li] markers): a condition can be a sub-case of the bullet above it. 4) For tables built from icons, read the aria-label (Checked/Unchecked, Included/Not included) instead of guessing from glyphs. 5) When a label is not in the text, open the screenshot itself to confirm it, and mark it as coming from a capture.

Adversarial fact-checking of research reports that cite knowledge-base pages, developer docs and pricing catalogs.