Full-text search the humor corpus. Returns lines with their credit attached. Filter by source pack, kind or minimum human score.

The highest human-rated lines in the corpus. Single-word lexicon entries use a different scale and are excluded unless kind='word'.

What this corpus's human rater actually liked versus disliked, with the score distribution. Use this to calibrate register before writing.

Structural breakdowns (mechanism, setup/turn, technique) for lines matching a query — how the joke is built, not just the text.

Chosen vs rejected pairs — what won head-to-head and what lost. Off-rubric packs are withheld by default and reported in withheld.

Every corpus pack loaded: who wrote it, under what license, whether it may be redistributed or used commercially, and how to cite it.

A compact, paste-ready style brief: exemplars, liked/disliked calibration, and the attribution block that must travel with any output.

What is loaded: counts by pack and by kind, size on disk.