Models trained to perform inner life score lowest; roleplay fine-tunes score below their own base models.

{
  "doc": "battery.md",
  "context": "Discriminant validity finding: Euryale (roleplay on Llama 70B) scores 1.81 vs base Llama 1.91. RP training suppresses self-observation.",
  "norm_label": "Models trained to perform inner life score lowest; roleplay fine-tunes score below their own base models.",
  "graphify_id": "finding_roleplay_suppression",
  "source_file": "battery.md",
  "imported_from": "/tmp/koan-debug/battery/graph.json",
  "extracted_type": "finding",
  "source_location": "Abstract, §3.2",
  "graphify_file_type": "finding"
}