Episode map
Each episode embedded from its full transcript and projected to 2D — nearby episodes cover similar ground. Color = the episode’s dominant theme. Click to open.
Topical range per episode
How far an episode's own chapters spread in embedding space (word-weighted mean distance to the episode's centroid, measured before 2D projection). For scale: the average distance to other episodes' chapters is 0.52 — every episode is far more cohesive than the corpus, but some roam much more than others.
Most wide-ranging
Most focused
Chapter map
All 885 substantive chapters, embedded and projected — the show’s conversational territory. Hover to identify; click to open that chapter's transcript. Highlight an episode above to see its footprint.
Notes on methodology
Every chapter is embedded locally (no API): a model2vec static text embedding of the chapter's title and text, blended with TF-IDF/SVD features that capture distinctive vocabulary. An episode's embedding is the word-weighted average of its chapters.
The 2D maps are t-SNE projections of those embeddings. Distances are meaningful locally — near neighbors talk about similar things — but the axes have no fixed meaning and long distances are distorted, so read clusters, not coordinates.
"Similar episodes" are nearest neighbors in the full embedding space, before projection. "Topical range" is the word-weighted mean distance from an episode's chapters to the episode's own centroid, also measured before projection. Colors group the 26 topics into 8 themes by each episode's or chapter's dominant topic.