The best teachers of almost any subject are on YouTube, free — and functionally unfindable. Search any topic and the ranking hands you engagement-optimized content: high-energy, low-density videos that feel educational while teaching four minutes of material per twenty. The genuinely great lecturers — dense, structured, pitched at a coherent level — rank below the entertainers because the algorithm measures retention, not learning.

Worse, self-learning needs a sequence: beginner material that doesn't condescend, intermediate content that assumes exactly what you now know, advanced treatment when you're ready. Assembling that path by trial and error costs weeks of watching mislabeled content — the "beginner" video requiring calculus, the "advanced masterclass" reteaching basics.

deep_search filters on the two axes the algorithm ignores. The level filter — transcript-classified as beginner, intermediate or advanced from what the content actually assumes — makes the sequence buildable. The high_density verdict separates teaching from padding: substance per minute, measured from the transcript, immune to production energy.

Combine with content types (lecture, explainer) and the niche-reach dial (the best teachers are often small channels the ranking buries), and curriculum-building becomes a filtered search per stage: dense beginner material first, dense intermediate next, advanced when you're ready. The path you assemble is graded by what's said, not by what thumbnails claim.