{
  "study": "long-life-20261118-r01-r08",
  "registered_revision": "4f5a2bf3e2331bde4b88f4d69917c6e1409e2a19",
  "original_archive_sha256": "ffecd50531eca3c1ec2442fe6ade54b02d62d87a1493ade202ea106c0f3b0eb1",
  "hardware": {
    "gpu": {
      "device": "NVIDIA RTX A5000",
      "sampled_device_capacity_mib": 24564,
      "peak_sampled_whole_device_memory_mib": 19029,
      "samples": 23332,
      "peak_first_sample_line": 2376,
      "source": "runtime-00/gpu-samples.log, comma-separated fields 2, 9 and 10 (one-based). The peak is a sampled whole-device reading."
    },
    "cpu": {
      "model": "12th Gen Intel(R) Core(TM) i7-12700K",
      "source": "runtime-00/ollama.log, line 29, runner device enumeration."
    },
    "host_memory": {
      "linux_memtotal_kib": 64262744,
      "gib_rounded": 61.29,
      "source": "runtime-00/before.json resources.memory.MemTotal and calls/0001.json resources.memory.MemTotal. This is OS-visible RAM."
    }
  },
  "inference": {
    "runtime": "Ollama",
    "generation_service_version": "0.40.15",
    "model_tag": "qwen3.8:27b",
    "model_digest": "aaee06c39dcf2437cde036998d960e1fc1494b8191be7cc9657d01e509097813",
    "model_format": "gguf",
    "parameter_size_reported": "27.3B",
    "quantization": "Q4_K_M",
    "native_packed_input_limit_tokens": 4096,
    "runner_context_tokens": 16384,
    "max_generated_tokens": 460,
    "temperature": 0.6,
    "thinking_mode_requested": false,
    "ollama_reported_size_bytes": 17361985862,
    "ollama_reported_size_vram_bytes": 17361985862,
    "placement_verified_on_calls": 4479,
    "source": [
      "runtime-00/runtime.json version and tags.models",
      "R01/A/config.json context_native_tokens and context_packing",
      "calls/0001.json through calls/4479.json placement, request.options and request.think"
    ]
  },
  "retrieval_and_storage": {
    "embedding_model": "embeddinggemma:300m",
    "embedding_model_digest": "85462619ee721b466c5927d109d4cb765861907d5417b9109caebc4e614679f1",
    "embedding_quantization_reported": "BF16",
    "dimensions": 768,
    "mode": "resident-scoped hybrid lexical and vector retrieval",
    "example_stored_vector_count": 496,
    "world_store": "SQLite",
    "source": [
      "runtime-00/runtime.json tags.models",
      "R01/A/world.sqlite memory_vectors.version and length(vector)",
      "village_zero/memory_index.py lines 15–16, 33–44 and 207–240 at the registered revision"
    ]
  },
  "usage": {
    "generation_calls": 4479,
    "registered_call_budget": 4992,
    "input_tokens": 17230730,
    "output_tokens": 768521,
    "model_prefill_plus_generation_seconds": 35129.705279566,
    "cumulative_request_seconds": 36658.510029,
    "monitor_start_to_final_update_seconds": 40338.916423,
    "first_request_to_final_reply_seconds": 40267.953373,
    "wall_time_through_publication_as_reported": "approximately 11 h 13 min",
    "transport_retries": 0,
    "transport_errors": 0,
    "decision_errors_recorded_by_monitor": 16,
    "source": [
      "statistics.json total_cost and cost_by_phase",
      "status.json started, updated and decision_errors",
      "calls/0001.json time through calls/4479.json response.created_at",
      "Long-Life Run results, lines 70–77 (wall time through archive publication)."
    ],
    "prompt_evaluation_seconds": 17511.040205629,
    "output_generation_seconds": 17618.665073937,
    "output_tokens_per_generation_second": 43.62,
    "output_rate_definition": "Sum of eval_count divided by sum of eval_duration, using only time spent generating output on these Qwen calls."
  },
  "measurement_limits": [
    "Counts and cumulative request timings cover Qwen generation. Embeddings, orchestration and publication contribute to elapsed time.",
    "Reported size_vram is the loaded runner allocation. The device monitor separately reports the card capacity and samples total device use.",
    "The GPU timing sums prompt evaluation and output generation reported by Ollama under full placement. Energy and whole-system operating costs would require additional metering.",
    "Throughput at larger populations and on other hardware remains an open question after this eight-history single-host experiment."
  ],
  "source_sha256": {
    "status.json": "cd62404c02c245b335dc5206e443b2667e8b462689aa277184d21dc6c265ecb0",
    "runtime-00/gpu-samples.log": "f87d7bc97c39b967676630b1f14d9c4ac20f7d9868eba60ba3b61fb4deeaed4a",
    "runtime-00/ollama.log": "68918038a449571f4b84afc691b63c917bfc200645ea545e61667a5c581f3f5a",
    "runtime-00/runtime.json": "57e1f800a0366ee8328b4b5b540935ec148c2d2d5efaec71bee371d1c6104c95",
    "runtime-00/before.json": "3b5a08e2cdcbc8cbb2939824f747413fc57cf1e653f3955d5af3cb48708f49a4",
    "statistics.json": "3615fbb4276deab65eb82adf96a85d910d78763251113143b1d96bf2a3d79696",
    "calls/0001.json": "0b1158a59f4283e46168dce4bd80784669aadfe0541fcd4487c0bb0a20b47391",
    "calls/4479.json": "5201cbc1b09d0f4f35047d6f7c81acafd37b1efc30e786d2e43f01c693e86ae4",
    "R01/A/config.json": "5038cba0a43e5dd358e95d04786075b54c20c27fb917eff85d44713e7bd7d1fa",
    "registration.json": "c7df25acc948731ef6ddbdda7d149a550d98877b87f9d1e9cfb1f2944ccc522f",
    "Long-Life Run results": "8c0964c6780fe67258e4caa38579695f78d8e3d05dbeeba687cfaa01366b831a"
  }
}
