{
  "same_requested_conditions": true,
  "summary": {
    "tacto": {
      "runs": 3,
      "passed": 3,
      "seconds": 126.29850741592236,
      "mean_seconds": 42.09950247197412,
      "total_input_tokens": 201447,
      "new_input_tokens": 22119,
      "cache_read_tokens": 179328,
      "output_tokens": 1605
    },
    "sky": {
      "runs": 3,
      "passed": 3,
      "seconds": 138.50689641595818,
      "mean_seconds": 46.16896547198606,
      "total_input_tokens": 433312,
      "new_input_tokens": 50592,
      "cache_read_tokens": 382720,
      "output_tokens": 1604
    }
  },
  "observed_reduction_percent": {
    "seconds": 8.814282404662432,
    "total_input_tokens": 53.509942027915216,
    "new_input_tokens": 56.27964895635673,
    "output_tokens": -0.06234413965087171
  },
  "basis": "three fresh independent paired case runs; requested gpt-6.1-sol/medium; fixture preparation excluded",
  "not_proven": [
    "actual serving-model identity",
    "server cache equality",
    "all-app generalization",
    "ten-job scripted-replay comparison"
  ],
  "source": "Local paired benchmark, 2026-10-14; docs/tacto-matched-comparison-2026-10-14.md",
  "working_prototype": true,
  "product": {
    "name": "Tacto",
    "measurement_label": "tacto"
  }
}