{
  "interpretability_model": "Qwen/Qwen2.5-1.5B-Instruct",
  "n_layers": 28,
  "best_probe_layer": 23,
  "best_probe_accuracy": 0.6514492753623189,
  "note": "This model is a dedicated white-box interpretability testbed, distinct from the 16-model black-box leaderboard."
}