A video in your niche just did twenty times the channel's usual numbers, and everyone has a theory: the thumbnail, the title, the algorithm's mood. Theories are free. What actually happened is measurable — which moments viewers replayed, what the comments latched onto, how the video opens versus how its channel usually opens — and almost nobody checks, because checking means forensic work per video.

Breakouts are the closest thing YouTube has to published experiments: some variable changed and the audience responded. Read enough of them properly and you stop cargo-culting thumbnails and start seeing the real patterns — which is exactly the homework worth doing before your next upload, on someone else's results.

run_youtube_tool runs the forensics on any public video, straight from a URL: most-replayed shows which segments viewers scrubbed back to — the actual hooks, not the assumed ones; comment-sentiment and comment-summarizer show what the audience responded to in their own words; hook-grader scores the opening against what makes openings work.

deep_search supplies the comparison set: the channel's ordinary videos and the niche's other breakouts, found at the transcript level. Run the same forensics across a handful and the pattern separates from the noise — was it the cold open, the topic's timing, the promise in the first fifteen seconds? You walk into your next script with evidence instead of superstition.