Characters Who Rate Below Their Own Castmates

Headline metric: vs castmates, with a 95% CI. For every episode a character appears in, we compare their average joke quality to the rest of the cast in that same episode, then average across episodes. Same LLM, same prompt, same show context. The ± value next to each delta is the 95% confidence interval half-width — only characters whose upper bound is still negative appear here. Small-sample flukes (Erin Hannon, Toby Flenderson) are automatically filtered out even though their point estimates are negative.

Avg and peak give more context: peak is the top 20% of their jokes. A character can trail castmates on average and still have elite peaks (Michael Scott: −0.126 vs cast, 6.69 avg, 7.83 peak — the high-volume-character archetype).

Straight men and plot-drivers typically trail the scene-stealers because they deliver the setups others punch off of. This measures the bits, not the character’s value to the show.