{
  "schema_version": 1,
  "release_status": "pilot-preview",
  "date": "2026-10-18",
  "title": "Business Analysis & Deliverables: Finance and Human Resources",
  "evidence_class": "curated-summary-of-local-trial-results",
  "runner": {
    "name": "Harbor",
    "version": "0.20.0"
  },
  "agent_harness": {
    "name": "Claude Code",
    "version": "2.1.114"
  },
  "protocol": {
    "provider": "Bailian Anthropic-compatible endpoint",
    "agent_model": "qwen3.7-max",
    "request_settings": {
      "thinking": "adaptive",
      "effort": "high",
      "max_tokens": 32000,
      "source": "observed Claude Code defaults"
    },
    "agent_runs_per_task": 1,
    "reference_runs_per_task": 1
  },
  "tasks": [
    {
      "id": "FIN-T1-003",
      "version": "1.1.2",
      "domain": "Finance",
      "inputs": 55,
      "source_type": "Public reference documents and synthetic business records",
      "outputs": [
        "docx",
        "xlsx"
      ],
      "graded_criteria": 25,
      "penalty_criteria_included": 5,
      "gating_criteria": 2,
      "positive_weight_sum": 172,
      "judge_model": "deepseek-v4.1-flash",
      "default_judge_model": "deepseek-flash"
    },
    {
      "id": "HR-T3-114",
      "version": "1.0.0",
      "domain": "Human Resources",
      "inputs": 17,
      "source_type": "Synthetic materials",
      "outputs": [
        "xlsx",
        "docx"
      ],
      "graded_criteria": 34,
      "penalty_criteria_included": 6,
      "gating_criteria": 0,
      "positive_weight_sum": 106,
      "judge_model": "qwen3.7-plus",
      "default_judge_model": "qwen3.7-plus"
    }
  ],
  "runs": [
    {
      "task_id": "FIN-T1-003",
      "task_checksum": "74a37955bebdc51000449bac2e119b19615c2d7179fd62635c582fb052e34d2d",
      "agent": "oracle",
      "agent_version": "1.0.0",
      "model": null,
      "status": "scored",
      "reward": 1.0,
      "graded_score": 1.0,
      "gating": 1.0,
      "verifier_error": 0.0,
      "criteria_counted": 25.0,
      "missing_outputs": [],
      "role": "reference",
      "evidence_sha256": {
        "result.json": "763659ba61acb2eea91ada7dd1c0eb28b995518f021d27810efee5da6eff1332",
        "reward.json": "3b5cf538afb08c4c06ceb638da311f9edc3aa36eb434adb04d17a3a300cd94a2"
      }
    },
    {
      "task_id": "FIN-T1-003",
      "task_checksum": "74a37955bebdc51000449bac2e119b19615c2d7179fd62635c582fb052e34d2d",
      "agent": "claude-code",
      "agent_version": "2.1.114",
      "model": "qwen3.7-max",
      "status": "scored",
      "reward": 0.332849,
      "graded_score": 0.332849,
      "gating": 1.0,
      "verifier_error": 0.0,
      "criteria_counted": 25.0,
      "missing_outputs": [],
      "role": "agent",
      "evidence_sha256": {
        "result.json": "67a9eb435a054bef5a569eeef90d4ec383952d98aaccc2eba2f0ec62d215b735",
        "reward.json": "bc1235c50cb110d000bf89eb21ce4c1c5586f888178ebb4002748138a3a20fb5"
      }
    },
    {
      "task_id": "HR-T3-114",
      "task_checksum": "7d047e2194a153d21fc1780c6e165346ad9cf827c09c3317627f8266671781bb",
      "agent": "oracle",
      "agent_version": "1.0.0",
      "model": null,
      "status": "scored",
      "reward": 0.783019,
      "graded_score": 0.783019,
      "gating": null,
      "verifier_error": 0.0,
      "criteria_counted": 34.0,
      "missing_outputs": [],
      "role": "reference",
      "evidence_sha256": {
        "result.json": "5506fd51ceed51695662da485cf47fb569a9e4dbfbed3438673a0000521bcecc",
        "reward.json": "293beec1cde595fdc0df17cff445d89f92a3c540043ce1f34d351b71a6786894"
      }
    },
    {
      "task_id": "HR-T3-114",
      "task_checksum": "7d047e2194a153d21fc1780c6e165346ad9cf827c09c3317627f8266671781bb",
      "agent": "claude-code",
      "agent_version": "2.1.114",
      "model": "qwen3.7-max",
      "status": "scored",
      "reward": 0.207547,
      "graded_score": 0.207547,
      "gating": null,
      "verifier_error": 0.0,
      "criteria_counted": 34.0,
      "missing_outputs": [],
      "role": "agent",
      "evidence_sha256": {
        "result.json": "2b84628c0b4a9fa32d8edc461febcdf983815aa73fc2e222f3b16a685daad63a",
        "reward.json": "6c9ea2bbd0b0cf6ca8027c9e27f9fb2edb07e2da0f7bcc9699feb6386a485032"
      }
    }
  ],
  "caveats": [
    "Two tasks with one agent run each; these are pilot observations, not official Challenge rankings or estimates of general performance.",
    "Task-specific rubric reward ranges from 0 to 1 and is not replication accuracy. Different tasks and judges should not be pooled into a headline score.",
    "HR reference scoring needs calibration: R21/R22 interpreted required details across both deliverables; R23 encountered missing cached spreadsheet formula values.",
    "verifier_error=1 denotes unavailable scoring, not a valid zero score. All four recorded trials have verifier_error=0.",
    "Final task release, competition rules and submission process are to be announced."
  ],
  "provenance": {
    "records_sha256": "227eebf8a3b8c65ebb806fd0a38d5cb226d6f807f59722f1820e464516b19d02",
    "method": "Summary fields checked against trial results and verifier rewards; original tasks and deliverables remain outside the public website."
  }
}