{
  "schema_version": "evalarc.harbor-import.v1",
  "source": {
    "result_sha256": "850450b131e1c2930e53a95497208f6a95a1362b7c5a161c4ab7b6d558461d05",
    "trajectory_sha256": null,
    "trial_name": "harbor-task__pYo9jgP",
    "task_name": "noteflowai/robot-evidence-review"
  },
  "upstream": {
    "rewards": {
      "reward": 1.0
    },
    "exception": null,
    "workflow_finished_at": "2026-09-14T10:59:44.651554Z",
    "agent_completion": null
  },
  "trajectory": null,
  "independent": {
    "evaluation": "evaluation.json",
    "candidate_sha256": "9a9922a8dfa396f483afd2228dfcd34a5ba78093dd6e14f1ab29422256cd09e1",
    "candidate_selection": "explicit caller-supplied directory",
    "valid": true,
    "resolved": false,
    "score": 0.625,
    "status": "failed"
  },
  "acceptance": {
    "minimum_independent_score": 1.0,
    "requires_valid_independent_evaluation": true,
    "requires_upstream_reward": false,
    "accepted": false
  },
  "scope": "Upstream rewards are imported claims. Acceptance uses the explicitly configured independent score threshold; resolved means every independent case passed. A finished Harbor workflow does not establish agent completion. ATIF inspection validates envelope and tool linkage, not the full upstream schema. The caller selects the candidate; source report metadata cannot establish that it is the same program used in the upstream run."
}