Parody post builds a Jev-style decision model in 25 lines of Python

Sources: primary, discussion
Summary: The code tokenizes each choice label, reads the logits for those token ids at the final position, and normalizes across them to get probabilities over a fixed option set. It runs against a local Qwen3-0.6B GGUF through llama-cpp-python. The author labels the post a parody and links OpenJev implementations for a complete version.
Why it matters: The runnable core shows how to constrain a local model to a fixed option set by normalizing over choice-token logits.