{
  "schema": "cain.e139.external_benchmarks.v1",
  "generated_at": "2026-11-21T22:27:27Z",
  "total_audited": 8,
  "completed": 0,
  "blocked": 8,
  "status": "BLOCKED",
  "benchmarks": [
    {
      "benchmark": "ARC-AGI-2",
      "source": "https://arcprize.org/arc-agi",
      "version": "ARC-AGI-2 (2025/2026)",
      "license": "Apache-2.0 public / private ARC Prize eval",
      "leakage_risk": "Public split widely leaked; only sealed private eval is uncontaminated",
      "participated": false,
      "status": "BLOCKED",
      "reason": "No private ARC Prize submission credentials or GPU frontier compute. Custom synthetic tasks are NOT ARC-AGI."
    },
    {
      "benchmark": "ARC-AGI-3",
      "source": "https://arcprize.org/",
      "version": "ARC-AGI-3 interactive preview",
      "license": "ARC Prize Foundation",
      "leakage_risk": "Low (interactive sealed environments)",
      "participated": false,
      "status": "BLOCKED",
      "reason": "Requires official interactive ARC-AGI-3 harness access."
    },
    {
      "benchmark": "METR Long-Horizon Autonomy",
      "source": "https://metr.org/",
      "version": "METR Time-Horizon Suite (2025/2026)",
      "license": "Restricted Vivaria suite",
      "leakage_risk": "Very low (private task bank)",
      "participated": false,
      "status": "BLOCKED",
      "reason": "Requires third-party METR auditor & cloud Vivaria VMs. Internal long-horizon runs are NOT METR scores."
    },
    {
      "benchmark": "SWE-bench Pro",
      "source": "https://www.swebench.com/",
      "version": "SWE-bench Pro (post-cutoff harder split)",
      "license": "MIT / Scale",
      "leakage_risk": "Low relative to SWE-bench Verified",
      "participated": false,
      "status": "BLOCKED",
      "reason": "Requires frontier coding model + Docker fleet; $0.00 paid API budget."
    },
    {
      "benchmark": "SWE-bench Verified (Historical Context Only)",
      "source": "https://www.swebench.com/",
      "version": "500-task Verified",
      "license": "MIT",
      "leakage_risk": "HIGH \u2014 documented pretraining contamination; MUST NOT be presented as an uncontaminated frontier measure",
      "participated": false,
      "status": "BLOCKED",
      "reason": "Prior historical self-run 257/500 is contaminated/interim only, NOT an official leaderboard submission."
    },
    {
      "benchmark": "LiveCodeBench (Post-Cutoff)",
      "source": "https://livecodebench.github.io/",
      "version": "v5/v6 (2026)",
      "license": "MIT",
      "leakage_risk": "Low (time-stamped post-cutoff)",
      "participated": false,
      "status": "BLOCKED",
      "reason": "Offline CPU host / $0.00 paid budget; internal 40-task suite is NOT LiveCodeBench."
    },
    {
      "benchmark": "CLADDER / CausalBench",
      "source": "Jin et al. / GSK",
      "version": "CLADDER v1",
      "license": "CC-BY / MIT",
      "leakage_risk": "Moderate",
      "participated": false,
      "status": "BLOCKED",
      "reason": "Offline host without external dataset download provisioned."
    },
    {
      "benchmark": "OSWorld / WebArena",
      "source": "OSWorld / CMU WebArena",
      "version": "2025/2026",
      "license": "Apache-2.0 / MIT",
      "leakage_risk": "Low",
      "participated": false,
      "status": "BLOCKED",
      "reason": "Requires multimodal agent + VM/Docker web stack."
    }
  ],
  "honest_disclosure": "0 external official benchmarks were completed in E139. SWE-bench Verified contamination is explicitly disclosed. Custom internal suites are NEVER presented as official ARC-AGI, METR, or SWE-bench scores."
}