{
  "schema_version": 1,
  "last_updated": "2026-05-15T00:00:00-04:00",
  "policy": "Render real benchmark values only when the row has a source URL, source date, and visible caveat. Otherwise render the benchmark as tracked or queued, not scored.",
  "benchmarks": [
    {
      "id": "metr-time-horizon",
      "label": "METR Time Horizon",
      "lane": "autonomy_horizon",
      "status": "live-primary",
      "source_url": "https://metr.org/time-horizons",
      "data_url": "https://metr.org/assets/benchmark_results_1_1.yaml",
      "source_updated": "2026-04-15",
      "display_rule": "Plot p50 and p80 human-expert minutes on a log scale. Do not project trend lines.",
      "caveat": "METR excludes central estimates above 16 hours from its doubling-time fit."
    },
    {
      "id": "epoch-ai-models",
      "label": "Epoch AI model database",
      "lane": "compute_frontier",
      "status": "source-linked",
      "source_url": "https://epoch.ai/data/ai-models",
      "source_updated": "2026-05-08",
      "display_rule": "Use for compute, cost, parameter, and release trend charts once the renderer pulls the CSV snapshot."
    },
    {
      "id": "lmarena",
      "label": "LMArena leaderboard",
      "lane": "open_frontier_proximity",
      "status": "source-linked",
      "source_url": "https://llmarena.ai/leaderboard",
      "display_rule": "Use current leaderboard state for frontier/open gap; label as rolling-state, not news."
    },
    {
      "id": "arc-agi-2",
      "label": "ARC-AGI-2",
      "lane": "capability_sota",
      "status": "queued-primary-audit",
      "source_url": "https://arcprize.org/leaderboard",
      "display_rule": "Render only after exact row, model, score, and date are captured."
    },
    {
      "id": "swe-bench",
      "label": "SWE-bench",
      "lane": "capability_sota",
      "status": "queued-primary-audit",
      "source_url": "https://www.swebench.com/",
      "display_rule": "Render verified/pro rows separately; do not merge suites without a formula."
    },
    {
      "id": "frontiermath",
      "label": "FrontierMath",
      "lane": "capability_sota",
      "status": "queued-primary-audit",
      "source_url": "https://epoch.ai/frontiermath",
      "display_rule": "Render only when public result rows include model, score, and date."
    },
    {
      "id": "robotics-endurance",
      "label": "Robotics endurance",
      "lane": "embodied_deployment",
      "status": "tracked-nonstandard",
      "source_url": "https://www.youtube.com/live/luU57hMhkak",
      "display_rule": "Track hours, intervention counts, recovery behavior, and task diversity only when directly visible or reported."
    }
  ]
}