{
  "lab_version": "1.0",
  "run_date": "2026-10-09",
  "environment": {
    "python": "3.13.5",
    "numpy": "2.3.5",
    "platform": "Linux-6.18.44-x86_64-with-glibc2.41",
    "device": "CPU"
  },
  "scope": "Random tiny model; numerical and state-management tests only. No API or pretrained-model benchmark.",
  "config": {
    "vocab": 41,
    "width": 32,
    "layers": 3,
    "q_heads": 4,
    "kv_heads": 2,
    "head_dim": 8,
    "seed": 7,
    "dtype": "float64"
  },
  "model_identity": "tiny-gqa-rope-v1:d71a873dd3c45a619daae621b2fb66f86897f5f6b16f068a738f212adb3e60f2",
  "prefix_tests": [
    {
      "dtype": "float64",
      "input_tokens": 47,
      "hit_tokens": 32,
      "computed_tokens": 15,
      "cached_vs_full_max_abs": 0.0,
      "split_vs_full_max_abs": 0.0,
      "cached_vs_split_max_abs": 0.0
    },
    {
      "dtype": "float32",
      "input_tokens": 47,
      "hit_tokens": 32,
      "computed_tokens": 15,
      "cached_vs_full_max_abs": 0.0,
      "split_vs_full_max_abs": 0.0,
      "cached_vs_split_max_abs": 0.0
    }
  ],
  "context_dependency": {
    "first_layer_document_k_max_abs": 0.0,
    "second_layer_document_k_max_abs": 2.112007194820824,
    "wrong_splice_next_logits_max_abs": 1.3119541099416636,
    "full_replay_repair_max_abs": 0.0
  },
  "online_softmax_max_abs": 5.551115123125783e-17,
  "quantization": {
    "format": "uniform signed INT4; FP32 scale per 16 values",
    "original_fp32_bytes": 8192,
    "hypothetical_fp16_bytes": 4096,
    "packed_data_bytes": 1024,
    "scale_bytes": 512,
    "total_payload_bytes": 1536,
    "reconstruction_max_abs": 0.21534329652786255,
    "next_logits_max_abs": 0.09491987950877667
  },
  "timing_note": "No latency benchmark: toy CPU timing is not evidence of GPU speedup."
}