One judge call vs twelve dimension scores

One direct Jev question per row against 12, 14 Jev-scored dimensions with locally fitted weights on three classification tasks: 5,477 test rows, 25,174 Jev calls, $1.43.

Decomposition wins on Japanese NLI (0.9076 vs 0.8373) but flags about 25× more hard benign rows as attacks (37.2% vs 1.5%). Reading queue from builders: essays, lists, and launch threads to supplement official docs. Tags in the wild: guides.

Follow the primary link for the author's latest notes and setup quirks.