Vulnerability × humor
Word-weighted mean of model-judged tone per episode. Hover any dot; a few standouts are labeled.
Warmest and darkest episodes
Mean chapter valence (−2 dark … +2 warm), top and bottom ten.
All episodes — tone table (sortable)
Notes on methodology
A language model (Claude Fable) read a digest of every transcript chapter and scored four tone dimensions:
- Valence — emotional warmth, from −2 (dark, pessimistic) to +2 (warm, optimistic).
- Vulnerability — from 0 (impersonal, analytical) to 3 (deeply personal, emotional disclosure).
- Humor — from 0 (serious) to 3 (frequently joking or laughing).
- Tension — from 0 (agreeable) to 3 (real disagreement or discomfort).
Episode-level numbers are word-weighted means of those chapter scores, with sponsor reads and intros ("meta" chapters) excluded. They are model judgments of transcript text — directional rather than precise, and they can miss delivery (a deadpan joke reads as serious on the page).
Valence is distinct from the sentiment curve on episode pages, which uses the VADER lexicon to score the word choice of each turn; valence is a judgment of a whole chapter's emotional register.