Results overview

Each model answers all questions twice: once as multiple choice (MCQ, graded automatically) and once open-ended (OEQ, graded by a PIAC-aware LLM judge that also flags hallucination). The gap between the two is the point.

By PIAC category

perceptual measurable from the signal · inferential trained analysis · affective subjective experience · contextual external world knowledge

Prompts

The exact prompt text driving every stage of the pipeline, pulled from the source modules.