According to @_avichawla, AnyJev extracts choices, scores, and probabilities from open causal LLMs with L0–L2 calibration for better decision outputs.