{
  "release": "Dot-v0.4-Thinking",
  "checkpoint_step": 938,
  "training_tokens": 8642015,
  "evaluation_scope": {
    "task_families": 8,
    "relationship_to_training": "same executable generators, separately seeded and hash-disjoint records",
    "warning": "This is targeted curriculum evaluation, not a broad reasoning or ARC-AGI benchmark."
  },
  "teacher_forced_4096": {
    "baseline_zero_gate": {
      "response_nll": 2.1283936659,
      "response_perplexity": 8.40136,
      "teacher_forced_token_accuracy": 0.666026,
      "teacher_forced_exact_response_rate": 0.0,
      "reasoning_nll": 2.511300,
      "reasoning_token_accuracy": 0.604705,
      "answer_nll": 0.0667919,
      "answer_token_accuracy": 0.9960501
    },
    "dot_v0_4": {
      "response_nll": 0.01736697,
      "response_perplexity": 1.017519,
      "teacher_forced_token_accuracy": 0.9942831,
      "teacher_forced_exact_response_rate": 0.8776855,
      "reasoning_nll": 0.0205677,
      "reasoning_token_accuracy": 0.9932305,
      "answer_nll": 0.00011697,
      "answer_token_accuracy": 0.9999480
    },
    "response_nll_change": -2.1110267
  },
  "free_running_256": {
    "thinking_envelope_valid_rate": 1.0,
    "nonempty_reasoning_rate": 1.0,
    "mean_reasoning_tokens": 23.8633,
    "mean_total_generated_tokens": 28.5977,
    "truncated_before_think_close_rate": 0.0,
    "final_answer_exact_match": 0.96875,
    "correct": 248,
    "examples": 256
  },
  "adversarial_spatial_probe_64": {
    "suite_relationship": "new spatial generators outside the v0.4 training curriculum",
    "final_answer_exact_match": 0.21875,
    "correct": 14,
    "examples": 64,
    "thinking_envelope_valid_rate": 0.875,
    "truncation_rate": 0.109375,
    "orientation_exact_match": 0.0,
    "warning": "The larger 256-case run was stopped after a 15-minute timeout caused by uncached verbose generation."
  }
}