Every getting-started tutorial is an involuntary confession. When a competitor's own video says "don't worry if the dashboard looks overwhelming at first," that's their onboarding admitting it overwhelms. When a third-party creator makes "the setup guide <rival> should have made," someone found the official path bad enough to fix for free. Minute-by-minute, tutorials document exactly where products need explaining — which is exactly where they failed to be self-explanatory.

Your rivals' confessions are all published. So are yours. A benchmark built from them answers the question your funnel data can't: not where users drop off, but what the drop-off feels like from the chair — and whether the competitor you're losing deals to has the same scar tissue or has actually solved it.

deep_search assembles the confession corpus: tutorial-type content for each rival product (and yours), both official and third-party — the third-party ones matter most, because nobody makes a workaround video for a flow that works. Evidence timestamps mark where each setup step gets explained.

library_import_youtube_video pins the shortlist into your library so the comparison has a stable base, and deep_context runs the side-by-side: your assistant lists the imported set, passes the item IDs, and asks like a UX researcher: where do tutorials slow down and over-explain? Which steps need a "common mistakes" section? What does the first ten minutes look like, product by product? Answers come back with citations to the exact moments, so the benchmark that lands in your onboarding review quotes the tape — including the uncomfortable clips about your own product.