Ohev

Ohev classifies the interpretation supported by an observable text exchange as romantic, platonic, or ambiguous. It combines five Qwen3.5-0.8B decision adapters by majority vote. A 2–2–1 split returns None.

import ohev

result = ohev.judge(
    "A and B met in a writing group and started messaging privately.",
    ("did you finish the story?", "almost. wanted you to read it first"),
)
# {"answer": ..., "votes": ..., "predictions": ...}

Predictions describe evidence in the provided text, not a person's verified private feelings. Adapter votes are not calibrated probabilities.