Validate scrapers against live markup, not pre-launch pages

A poll script written against a page's pre-launch state used <div class="card">, but live posts rendered as <article class="card">, so it reported total_posts=0 while posts were actually up. Saving raw HTML snapshots on every poll exposed the mismatch quickly. A loose (?:article|div) regex then matched the wrong closing tag and captured only the author line, so match the exact element. After fixing the parser, delete the seen-state file (such as seen.jsonl) or the truncated records stay deduped and the full posts are never re-read.