{
  "description": "Completed experiments used to illustrate a proposed ground-truth and reference-execution benchmark. Not a live evaluation service.",
  "token_accounting": "Accumulated run total tokens, including repeated context; not monetary cost.",
  "aggregation": "Time and tokens are per-run medians. Do not pool success rates across experiments.",
  "hn_comparability": "Live runs at different times and different completion volumes: CU verified 220 comments; WebMCP 282. Not an equal-work speedup comparison.",
  "experiments": [
    {
      "name": "Books to Scrape",
      "computer_use": {
        "successful_runs": 5,
        "runs": 5,
        "median_wall_seconds": 46.807,
        "median_total_tokens": 92559
      },
      "webmcp": {
        "successful_runs": 5,
        "runs": 5,
        "median_wall_seconds": 29.855,
        "median_total_tokens": 44440
      }
    },
    {
      "name": "X",
      "computer_use": {
        "successful_runs": 5,
        "runs": 5,
        "median_wall_seconds": 62.176,
        "median_total_tokens": 134038
      },
      "webmcp": {
        "successful_runs": 5,
        "runs": 5,
        "median_wall_seconds": 55.748,
        "median_total_tokens": 145301
      }
    },
    {
      "name": "Hacker News",
      "computer_use": {
        "successful_runs": 0,
        "runs": 3,
        "median_wall_seconds": 968.067,
        "median_total_tokens": 9582590
      },
      "webmcp": {
        "successful_runs": 3,
        "runs": 3,
        "median_wall_seconds": 407.268,
        "median_total_tokens": 1184254
      }
    }
  ]
}