{
  "benchmark": "Deep-SWE (Datacurve)",
  "tasks": 113,
  "attemptsPerTask": 3,
  "attemptsPerSetup": 339,
  "model": "GPT-5.6 Terra",
  "reasoning": "medium",
  "solveDefinition": "All required new and regression tests pass under the official verifier.",
  "memtrace": {
    "solved": 132,
    "genuineScored": 339,
    "cohort": "bestsetup-full339-20261021",
    "source": "benchmark-deepswe/bestsetup-full339-20261021/analysis/RESULTS.json",
    "sourceSha256": "463217872065ed2a8aae7d6e8f99c940a3fda03dafe32cf696b0c555596f5149"
  },
  "ordinary": {
    "solved": 112,
    "genuineScored": 339,
    "cohort": "full4arm1",
    "source": "benchmark-deepswe/full4arm1/reports/RESULTS-20261020T0241Z.json",
    "sourceSha256": "b546175c79d8c82b59ded14867aa51aafcf20e7618fd4f3c29779387c9ab431f"
  },
  "comparison": "Best fully scored Memtrace run compared with an earlier ordinary-tools baseline. These are separate runs with different harness configurations.",
  "upliftDefinition": "(Memtrace solves - ordinary solves) / ordinary solves * 100. Relative improvement, not a percentage-point increase."
}