{
  "created_utc": "2026-09-23T10:26:41.743208+00:00",
  "sample_count": 100,
  "diagnostic_sample": "-tPCytz4rww/12",
  "seed": 42,
  "device": "cuda",
  "gpu_name": "NVIDIA GeForce RTX 5070 Ti",
  "python": "3.14.7",
  "torch": "2.14.0+cu130",
  "experiments": [
    {
      "name": "D0",
      "scope": "single sample; span + barycenter band only",
      "steps_per_model": 1000,
      "loss": "5 * L_span + 10 * L_band"
    },
    {
      "name": "D1",
      "scope": "single sample; Gaussian target KL only",
      "steps_per_model": 1000,
      "sigma_normalized_time": 0.1,
      "M4_control": "no PE versus fixed sinusoidal PE"
    },
    {
      "name": "D2",
      "scope": "all 100 clips; Gaussian target KL only; one seed; in-sample diagnostic",
      "steps_per_model": 500,
      "sigma_normalized_time": 0.1,
      "M4_position_encoding": "fixed sinusoidal"
    },
    {
      "name": "D3",
      "scope": "all 100 clips; Gaussian KL + masked reconstruction + contrastive; one seed; in-sample diagnostic",
      "steps_per_model": 500,
      "sigma_normalized_time": 0.1,
      "M4_position_encoding": "fixed sinusoidal"
    }
  ],
  "optimizer": "AdamW",
  "learning_rate": 0.001,
  "dropout": 0.0,
  "batch_size": 8,
  "gradient_logging_interval_steps": 20,
  "gradient_metrics": [
    "W_Q",
    "W_K",
    "M4 latent slots Z"
  ],
  "features_changed": false,
  "M1_M2_changed": false,
  "elapsed_seconds": 103.73381090164185,
  "interpretation_limits": [
    "D0 and D1 overfit one selected sample and diagnose optimization/representability only.",
    "D2 and D3 train and evaluate on the same 100 clips; they diagnose whether the target can be optimized, not generalization.",
    "Gaussian targets are weak temporal priors constructed from timestamps; they are not human alignment ground truth.",
    "M4 absolute sinusoidal encoding is enabled only for D1's PE control and D2/D3; prior M1-M4 and v2 results are unchanged."
  ]
}